UK BUSINESS AI UNIVERSE: Places: London: Labs

AI activity, tracked by AI

104 records · Labs

by Alex Connock

Experimental site researched using Gen AI.
Based on public sources. Hallucination & omission likely.

BETAwww.librairie.ai

← Back to the Places: London tiles

Labs

104 records Generated from this page's data. Click a name to open its panel on the Places: London page. Records resting on one source are tagged "one source".
RecordWhat it isPlaceLinksConfidence
Aardvark Weather (end-to-end AI weather prediction)Chips, models and productsAI system that goes straight from raw weather observations to forecasts, replacing the traditional forecasting pipeline with one model. research (Nature paper)Key facts Published in Nature vol. 641, issue 8065, pp. 1172–1179 (2025); built by the Alan Turing Institute with the University of Cambridge and Microsoft Research; local station forecasts skilful up to 10 days ahead; Turing states it is "tens of times faster" and uses "thousands of times less computing power", running on a regular desktop computer; access: paperMade by The Alan Turing Institute (AI assurance and security work) · 2025 (Nature vol. 641; Turing blog 20 Mar 2025)Londonturing.ac.uk/science-innovation/environment-and- · turing.ac.uk/news/publications/end-end-data-drivlab's own pages
AI Research Centre for Defence (ARC-D), Dstl and The Alan Turing InstituteSchemes, programmes and AI linesA London centre set up by Dstl and the Alan Turing Institute to do defence AI research with UK universities. launched (current status not verified)Key facts First named the Defence Centre for AI Research (DARC); expected to create at least 5 to 8 new academic posts from September 2022; funded through the MOD Chief Scientific Adviser (amount not published); research themes: learning from little data, AI for war gaming, limits of AI models, managing multiple sensors, human-centric AI, ethics (all as of 2022-07-14)Londongov.uk/government/news/launching-the-defence-cenone source
ALOHA Unleashed and DemoStartChips, models and productsTwo robot-learning methods: one teaches two-armed robots complex tasks, the other trains dexterous hands in simulation. research (paper only)Key facts Announced 12 Sep 2024; DemoStart needs 100x fewer simulated demonstrations; over 98% success in simulation on several tasks; real-world 97% on cube reorientation and lifting, 64% on plug-socket insertion; access: paper onlyMade by Google DeepMind · 2024-09-12Londondeepmind.google/blog/advances-in-robot-dexterityone source
AlphaCode 2Chips, models and productsSecond AlphaCode system for competitive programming, solving more contest problems than the first. research (technical report)Key facts Solved 1.7x more problems than AlphaCode; performed better than 85% of competition participants; technical report not opened (storage.googleapis.com blocked); access: paper onlyMade by Google DeepMind · 2023 (reported in DeepMind review dated 22 Dec 2023)Versions AlphaCode → AlphaCode 2Londondeepmind.google/blog/2023-a-year-of-groundbreakione source
AlphaEvolveChips, models and productsGoogle DeepMind's coding agent that uses Gemini models plus automatic checkers to discover and improve algorithms. released (early access for academics)Key facts Recovers about 0.7% of Google's worldwide compute via a Borg scheduling heuristic; 4x4 complex matrix multiplication in 48 scalar multiplications; 23% faster Gemini kernel, 1% less Gemini training time; tried on 50+ open maths problems, matched best known in ~75%, improved in ~20% (all 14 May 2025). UK scientists offered priority access under the Dec 2025 UK government partnershipMade by Google DeepMind · 2025-05-14Londondeepmind.google/discover/blog/alphaevolve-a-gemi · deepmind.google/blog/strengthening-our-partnershlab's own pages
AlphaGeometry 2Chips, models and productsImproved version of AlphaGeometry, pairing a Gemini-based language model with a symbolic engine to solve olympiad geometry. research (competition result)Key facts Announced 25 Jul 2024; solved 83% of IMO geometry problems from the past 25 years (predecessor 53%); proved IMO 2024 Problem 4 within 19 seconds of receiving its formalisation; access: paper onlyMade by Google DeepMind · 2024-07-25Versions AlphaGeometry → AlphaGeometry 2Londondeepmind.google/blog/ai-solves-imo-problems-at-sone source
AlphaProofChips, models and productsReinforcement-learning system that proves maths statements in the formal language Lean; solved three IMO 2024 problems. research (paper; competition result)Key facts Announced 25 Jul 2024; with AlphaGeometry 2 solved 4 of 6 IMO 2024 problems, 28 of 42 points (silver-medal standard); AlphaProof solved two algebra and one number theory problem; methodology published in Nature 12 Nov 2025 (s41586-025-09833-y); access: paper onlyMade by Google DeepMind · 2024-07-25Londondeepmind.google/blog/ai-solves-imo-problems-at-sone source
Applied ComputingCompaniesLondon company building AI foundation models for oil, gas and petrochemical operations, sold through a platform called Orbital.
Profile
Founders
Callum Adamson — CEO and co-founder (other founders not named)
Chief executive
Callum Adamson — from null
Chief scientist or research lead
Sam Tukra — Chief AI Officer
Other leaders
Dan Jeavons — President (former Shell AI lead; named July 2025 per uktech.news link headline)
Hq
London (England: London); other_offices: Bengaluru; Houston, Texas (new office funded by the 2026 round)
What it does
A London company building Orbital, which it calls the first foundation model for energy operations. Orbital combines physics-based AI with models for chemical engineering, time-series forecasting and language, so oil, gas and petrochemical engineers can design and run plants more efficiently and with lower emissions.
Research focus
physics-grounded AI; multi-foundation models for energy and process industries
Models and products
Orbital — "multi-foundation model" for energy operations — 2025–26
Customers or partners named
KBR — investor and multi-year agreement to build exclusive AI products for energy — 16 July 2026 — https://www.uktech.news/ai/applied-computing-raises-14-8m-to-scale-ai-for-energy-operators-20260716; Databricks Ventures — investor — 16 July 2026 — https://tech.eu/2026/07/16/applied-computing-lands-20m-to-expand-foundation-ai-for-energy/
Funding total as published
null (2025 seed £9m and 2026 Series A $20m reported separately)
Rounds found
2025 (techmarketview link dated 2 June 2025; deal file has 2025-05-30) — seed — 9,000,000 GBP — (already DEAL-SU-ENR-020); 2026-07-16 — Series A — 20,000,000 USD (£14.8m) — KBR — Databricks Ventures (already DEAL-LAB-037 / DEAL-SU-ENR-008)
Status now
Active London company; raised a $20m Series A led by KBR in July 2026 and is opening a Houston office.

Labs Round 7 deep profile (30 Oct 2026), as the researchers wrote it; confidence: single_source (three trade sources; appliedcomputing.com blocked).

Sources

  1. tech.eu/2026/07/16/applied-computing-lands-20m-t
  2. uktech.news/ai/applied-computing-raises-14-8m-to
  3. techmarketview.com/news/archive/2026/07/23/ukhot
Models and products (1)
  1. Orbital (Applied Computing energy foundation model) — (flagship product named in release of 28 May 2025; first public date not stated) live (product)Applied Computing's AI for oil refineries and petrochemical plants, combining plant sensor data, physics and a chemical-engineering language model.
London (EC2A 4NE)Companies House · tech.eu/2026/07/16/applied-computing-lands-20m-t · find-and-update.company-information.service.gov. · find-and-update.company-information.service.gov. · find-and-update.company-information.service.gov.one source
Basecamp ResearchCompaniesCollects genetic data from nature and trains biological foundation models, called EDEN, to design gene therapies, enzymes and peptides.
Profile
Founders
Glen Gowers — co-founder and CEO; Oliver Vince — co-founder
Chief executive
Glen Gowers — from null
Chief scientist or research lead
John Finn — Chief Scientific Officer; Phil Lorenz — Chief Technology Officer
Other leaders
Richard Pearce — Chief Business Officer (formerly Biogen); Emma Bolton — VP of Global Research; Nick Coleman — Finance Director; Konstantinos Andrikopoulos — Senior VP of IP; Sarah Cross — VP of People; Andy Conrad (S32) — board member (Sep 2026)
Hq
London (England: London) — Clerkenwell Workshops, EC1R 0AT; other_offices: Boston, with labs at LabCentral, Cambridge, Massachusetts
What it does
Collects genetic data from nature with partners in more than 30 countries, builds a proprietary dataset (BaseData), and trains biological foundation models (EDEN) on it. It uses EDEN to design therapeutics and is advancing a pipeline of them towards clinical development.
Research focus
biological foundation models trained on evolutionary genomic data; AI-designed therapeutics; programmable gene insertion
Models and products
EDEN — biological foundation models; next generation funded by Series C — Sep 2026; Trillion Gene Atlas — dataset expansion — 18 Mar 2026; BaseData — proprietary evolutionary genomic dataset
Customers or partners named
NVIDIA, Anthropic (incl. Claude Science), PacBio, Ultima Genomics — partners — 23 Sep 2026 — https://www.basecamp-research.com/news-research/basecamp-research-raises-140m-to-advance-ai-designed-therapeutics; Microsoft Azure — EDEN training — 19 Aug 2026 — https://www.basecamp-research.com/news-research; Nebius — EDEN trained 20x faster — 23 Jun 2026 — same url; AI Biosecurity Lab (with Anthropic, OpenAI, DeepMind) — joined — 24 Oct 2026 — same url
Compute
Microsoft Azure (Aug 2026) and Nebius (Jun 2026) named for EDEN training (news titles)
Valuation reported
800,000,000 USD — 23 Sep 2026 — Reuters headline as listed on https://www.basecamp-research.com/news-research (Reuters page could not be opened; single_source)
Rounds found
2026-09-23 — Series C — 140,000,000 USD — S32 — Menlo Ventures' Anthology Fund, Catalio Capital Management, European Tech Collective, Firebrand River Capital, Inception Fund, King Philanthropies, NATO Innovation Fund, NVIDIA, PostScriptum, Redalpine, The Rockefeller Foundation, Singular, Sovereign AI, True Ventures; André Hoffmann among individual investors
Uk government links
Sovereign AI invested in Series C (Sep 2026); member of the AI Biosecurity Lab (Oct 2026; government funding held as DEAL-LAB-2021)
Status now
Raised a $140m Series C in September 2026 to train new EDEN models and take AI-designed therapeutics towards the clinic.

Labs Round 7 deep profile (30 Oct 2026), as the researchers wrote it; confidence: single_source.

Sources

  1. basecamp-research.com/news-research/basecamp-res
  2. basecamp-research.com/about
  3. basecamp-research.com/news-research
Models and products (4)
  1. Trillion Gene Atlas 2026-03-18 announced (data programme)Basecamp Research's plan to sequence a trillion genes from nature to train its AI models.
  2. EDEN (family of biological foundation models) 2026-01-12 released (two capabilities live; chat connector in Claude)Basecamp Research's 28-billion-parameter DNA foundation model family for designing programmable therapies, including gene-insertion enzymes.
  3. aiPGI (Basecamp Research) 2026-01-12 announcedBasecamp Research's AI gene-insertion platform, powered by its EDEN models, for designing gene-editing enzymes.
  4. HiFi-NN 2025-06-20 released (paper)A Basecamp Research model for labelling the functions of unknown microbial proteins ("microbial dark matter").
Clerkenwell Workshops, 27 Clerkenwell Close, London (EC1R 0AT)Companies House · basecamp-research.com/news-research/basecamp-res · basecamp-research.com/about · find-and-update.company-information.service.gov. · blogs.nvidia.com/blog/uk-ecosystem-deployment/verified
BeyondMathCompaniesLondon company building a generative physics model that produces engineering simulations quickly for engineering teams.
Profile
Founded
2022
Founders
Alan Patterson — co-founder and CEO; Darren Garvey — co-founder
Chief executive
Alan Patterson — from null
Hq
London (England: London); other_offices: null
Headcount
null (plans to double headcount in 2026, per CIC)
What it does
A London company building a generative "foundational physics" AI model trained on first-principles physics. Engineers use it to run simulations of airflow, heat and similar problems much faster than conventional simulation. It works with customers in cars, aerospace, electronics, data-centre design and chip-making, including a Formula 1 team.
Research focus
physics foundation models; engineering simulation (aerodynamics, thermal management)
Models and products
foundational physics AI model (unnamed) — 2026
Customers or partners named
Honeywell — STRATA, a "$19 million, 3-year project" — reported 25 February 2026 — https://www.cic.vc/beyondmath-completes-18-5-million-seed-round-to-scale-worlds-largest-foundational-physics-ai-model/; NVIDIA and AWS — "established partnerships" — 25 February 2026 — same URL
Compute
claims results "up to 1000x faster than traditional supercomputing methods" — February 2026
Funding total as published
18,500,000 USD (seed round total; £13.8m) — as of 25 February 2026 — https://www.cic.vc/beyondmath-completes-18-5-million-seed-round-to-scale-worlds-largest-foundational-physics-ai-model/
Rounds found
2024-08-20 — seed — 8,500,000 USD — UP.Partners (already DEAL-LB-132); 2026-02-25 — seed extension — 10,000,000 USD (£7.4m) — Cambridge Innovation Capital — UP.Partners, Insight Partners, InMotion Ventures (already DEAL-LAB-034)
Status now
Active London company; closed its seed round at $18.5m in February 2026.

Labs Round 7 deep profile (30 Oct 2026), as the researchers wrote it; confidence: verified.

Sources

  1. cic.vc/beyondmath-completes-18-5-million-seed-ro
  2. cic.vc/beyondmath-completes-18-5-million-seed-ro
  3. uktech.news/ai/beyondmath-raises-seed-extension-
LondonCompanies House · uktech.news/ai/beyondmath-raises-seed-extension- · find-and-update.company-information.service.gov. · find-and-update.company-information.service.gov. · find-and-update.company-information.service.gov.verified
BoltzCompaniesBuilds biomolecular models and AI agents that predict molecular structures and help design medicines; founded by MIT researchers.
Profile
Legal name
Boltz PBC (public benefit corporation; "Boltz, PBC" in releases)
Founded
null (founded by MIT researchers)
Founders
Gabriele Corso — co-founder and CEO (other founders not named on fetched pages)
Chief executive
Gabriele Corso — from null
Hq
Cambridge, Massachusetts (dateline of company releases) (Outside the UK); other_offices: London (described as "London-based" by NVIDIA, 13 October 2026; no London office on company pages)
Headcount
null (launch post names 12 team members, January 2026)
What it does
A public-benefit company building open-source and commercial AI models that predict and design proteins and small molecules for drug discovery and other chemistry. Its open models (Boltz-1, Boltz-2, BoltzGen) are widely used, and it sells Boltz Lab and an API to drug companies.
Research focus
biomolecular structure prediction; binding affinity; protein and small-molecule design; AI agents for scientists
Models and products
Boltz-1 — open biomolecular interaction model — 16 November 2024; Boltz-2 — binding affinity prediction — 6 June 2025; BoltzGen — universal binder design — 26 October 2025; Boltz Lab — platform (preview/beta) — 8 January 2026; BoltzMol-1, BoltzProt-1 and Boltz API — 16 June 2026; Claude Science integration — 30 June 2026
Customers or partners named
Pfizer — multi-year strategic collaboration; Pfizer keeps ownership of compounds — 8 January 2026 — https://boltz.com/pfizer-partnership; Takeda — collaboration to deploy models — 18 June 2026 — https://boltz.com/news; GSK — collaboration to deploy models — 2 July 2026 — https://boltz.com/news; dsm-firmenich — partnership on AI for ingredient discovery — 14 September 2026 — https://boltz.com/news; Bristol Myers Squibb — agreement to deploy models and co-develop custom models — 21 October 2026 — https://boltz.com/bms-partnership
Papers and awards
technical posts on Boltz-1 (2024), Boltz-2 (2025), BoltzGen (2025); open models "used by over a hundred thousand scientists" (January 2026)
Funding total as published
28,000,000 USD (seed) — as of 8 January 2026 — https://boltz.com/launch; NVIDIA planned investment October 2026, amount not disclosed
Rounds found
2026-01-08 — seed — 28,000,000 USD — lead not stated — Zetta, Amplify, a16z, Factorial Capital, Obvious Ventures, angels including Clement Delangue (NEW: DEAL-LAB-9005); 2026-10-13 — strategic investment (planned) — undisclosed — NVIDIA (already DEAL-LAB-2001)
Status now
Active; signed a Bristol Myers Squibb agreement on 21 October 2026; NVIDIA announced a planned investment on 13 October 2026.
Legal name (Labs Round 8)
Not stated (Boltz is a public-benefit corporation, "Boltz PBC")
Founded (Labs Round 8)
Not stated (team began as AI researchers at MIT; company announced 8 Jan 2026)
Founders (Labs Round 8)
Gabriele Corso — co-founder and CEO; Jeremy Wohlwend — co-founder; Saro Passaro — co-founder
Chief executive (Labs Round 8)
Gabriele Corso
Other leaders
Not stated (launch page lists team members Francesco Capponi, Demitri Nava, Hannes Stark, Noah Getz, Talip Ucar, Luca Cavalleri, Simone Primarosa, Geoffrey Smith, Luis Sanmiguel by name only)
Hq (Labs Round 8)
Not stated (all releases datelined Cambridge, MA — Outside the UK); other_offices: UK office not confirmed in any page opened
Headcount (Labs Round 8)
Not stated (12 team members named on 8 Jan 2026 launch page)
What it does (Labs Round 8)
Boltz builds open and commercial AI models that predict how biological molecules fit together and design new drug molecules and binders. It sells them through Boltz Lab and an API to drug companies including Pfizer, Takeda, GSK and Bristol Myers Squibb.
Research focus (Labs Round 8)
"Boltz is a frontier research lab building generative models for biology and chemistry."
Models and products (Labs Round 8)
Boltz-1 — open biomolecular interaction model — 16 Nov 2024; Boltz-2 — binding-affinity prediction model — 6 Jun 2025; BoltzGen — universal binder design — 26 Oct 2025; Boltz Lab — design platform — 8 Jan 2026; BoltzMol-1 — small-molecule hit discovery pipeline — 16 Jun 2026; BoltzProt-1 with Boltz-PPI — protein design pipeline — 16 Jun 2026; Boltz API — 16 Jun 2026
Customers or partners named (Labs Round 8)
Pfizer — strategic collaboration — 8 Jan 2026 — https://boltz.com/news; Takeda — collaboration — 18 Jun 2026 — https://boltz.com/news; GSK — collaboration incl. fine-tuned exclusive models — 2 Jul 2026 — https://boltz.com/gsk-partnership; dsm-firmenich — ingredient discovery — 14 Sep 2026 — https://boltz.com/news; Bristol Myers Squibb — deployment on BMS NVIDIA DGX SuperPOD and co-developed models — 21 Oct 2026 — https://boltz.com/bms-partnership; launch partners for API incl. Benchling, Amazon Bio Discovery, Rowan, Tamarind — 16 Jun 2026 — https://boltz.com/boltzmol-boltzprot-api
Selected papers
Introducing Boltz-1 — boltz.com — 16 Nov 2024 — https://boltz.com/boltz1; Introducing Boltz-2 — boltz.com — 6 Jun 2025 — https://boltz.com/boltz2; Introducing BoltzGen — boltz.com — 26 Oct 2025 — https://boltz.com/boltzgen
Funding total as published (Labs Round 8)
$28m seed — as of 8 Jan 2026 — https://boltz.com/launch
Rounds found (Labs Round 8)
8 Jan 2026 — seed — $28m — Amplify, a16z, Zetta (per page metadata) — Clement Delangue, Factorial Capital, Obvious Ventures (already DEAL-LAB-9005); 13 Oct 2026 — NVIDIA (already DEAL-LAB-2001; not seen this round)
Status now (Labs Round 8)
Trading; signed a deal with Bristol Myers Squibb on 21 October 2026, its fifth named pharma or industry partner in 2026.

Labs Round 7 and Labs Round 8 deep profile (30–31 Oct 2026), as the researchers wrote it; confidence: one source (company only); UK base unconfirmed.

Sources

  1. boltz.com/launch
  2. boltz.com/bms-partnership
  3. boltz.com/pfizer-partnership
  4. boltz.com/news
  5. boltz.com/
  6. blogs.nvidia.com/blog/uk-ecosystem-deployment/
  7. techfundingnews.com/nvidia-pledges-250m-of-compu
  8. boltz.com/platform
  9. boltz.com/manifesto
  10. boltz.com/boltzmol-boltzprot-api
Models and products (4)

Version chainsBoltz-1 → Boltz-2 → BoltzGen

  1. Boltz-2 (Boltz) 2026-06-16 released (open source)Open model that predicts both protein structure and how strongly a drug molecule binds.
  2. BoltzMol-1, BoltzProt-1 and the Boltz API (Boltz) 2026-06-16 liveBoltz's small-molecule and protein design pipelines, sold through a pay-per-prediction API.
  3. BoltzGen (Boltz) 2025-10-26 released (MIT licence)Open generative model that designs new proteins and peptides to stick to chosen disease targets.
  4. Boltz-1 (Boltz) 2024-11-16 released (MIT licence)Open-source model that predicts the 3D shape of proteins and the molecules that bind to them.
Londonblogs.nvidia.com/blog/uk-ecosystem-deployment/ · tech.eu/2026/10/15/nvidia-bets-on-five-uk-startu · boltz.com/launch · boltz.com/bms-partnershipverified
Centre for Long-Term Resilience (CLTR)Networks, bodies, events and mediaA London think tank that researches extreme risks from AI and biology and advises the UK government.Key facts Non-profit company registered in England and Wales; focus areas AI risk, biological risk, government risk management; 2026 outputs include "Priorities for the UK Government on frontier AI risk and Biosecurity" (2026-09-09), "AI loss of control incidents are worsening, shows CLTR analysis" (2026-08-29), "AI and the systemic management of risk – what the frontier labs can learn from business" (2026-09-24); relies on voluntary donations and takes no government funding (read 2026-10-31)Londonlongtermresilience.org/ · longtermresilience.org/the-team/one source
CodeGemmaChips, models and productsSmall open Gemma models trained to complete, write and chat about computer code. released (open weights)Key facts Initial release 9 Apr 2024; v1.1 released 3 May 2024; variants: 7B pretrained, 7B instruction-tuned, 2B pretrained (stated "up to 2x faster code completion"); trained on 500 billion tokens; access: open weightsMade by Google DeepMind · 2024-04-09Versions Gemma → Gemma 2 (and CodeGemma, RecurrentGemma, PaliGemma → PaliGemma 2, ShieldGemma → ShieldGemma 2, DataGemma)Londonai.google.dev/gemma/docs/codegemma · ai.google.dev/gemma/docs/releaseslab's own pages
CogitatCompaniesTrains its own AI/ML brainwave decoders on EEG data for non-invasive brain-computer interfaces.
Models and products (1)
  1. Cogitat brainwave decoders 2026-10-31 (as read) offered (as read)Cogitat's AI decoders that turn brain signals from non-invasive EEG headsets into digital commands, built to work across devices without long calibration.
London (registered office W8 6SN)Companies House · cogitat.io/verified
Concierge Chat (Omnichannel Intelligence Hub)Chips, models and productsPolyAI's chat assistant for retail websites, replacing search boxes with a single guided conversation. live (product)Key facts Post 7 Jan 2026; for one retail client over Black Friday weekend: about 27,000 conversations, over 81,000 dialogue turns, 73.2% handled without a human, about 2,250 staff hours saved (company figures); access: sales-ledMade by PolyAI · 2026-01-07Londonpoly.ai/blog/how-polyai-brought-its-omnichannel-one source
ConcrCompaniesCambridge University spin-out that uses astrophysics-derived Bayesian AI "digital twins" to predict how cancer patients respond to treatment.Key facts £1.94m seed 2023 co-led by Cambridge Enterprise Ventures and R42 Group; £0.8m Innovate UK grant for clinical validation; first patients recruited to a breast-cancer observational trial July 2024; 2026-09-03 pancreatic-cancer study in Frontiers in Artificial Intelligence: AUC 72.3% vs 44.8% for conventional biomarkers
Models and products (1)
  1. FarrSight (FarrSight-Twin) (Concr) 2026-09-03 research use onlyConcr's Bayesian AI model that builds a "digital twin" of a patient's tumour to predict treatment response.
London (described as "London-based" in trade; Cambridge Enterprise portfolio)Companies House · enterprise.cam.ac.uk/?p=33665 · precisionmedicineonline.com/precision-oncology/cverified
ConjectureCompaniesLondon AI research company building large language models and a new AI design meant to keep advanced AI controllable and auditable.
Profile
Founded
March 2022
Founders
Connor Leahy — co-founder; Sid Black — co-founder; Gabriel Alfour — co-founder (titles not given on the fetched pages)
Hq
London (England: London); other_offices: null
Headcount
20 — 15 February 2024 — https://www.conjecture.dev/research/conjecture-2-years
What it does
A London AI safety company that grew out of the open-source group EleutherAI. It started as an alignment research start-up and now describes its work as building large language models and "Cognitive Emulation" systems, meant to be bounded, controllable and auditable.
Research focus
AI control; Cognitive Emulation (CoEm); alignment; interpretability (e.g. "Interpreting Neural Networks through the Polytope Lens")
Models and products
LLM suite and Cognitive Emulation approach — as stated on About page, date null
Customers or partners named
null ("first enterprise customers" mentioned, unnamed, February 2024)
Papers and awards
research posts on conjecture.dev (e.g. "Cognitive Emulation: A Naive AI Safety Proposal") — dates not checked
Rounds found
2022-04-08 (announcement) — seed-stage (stage not stated) — amount not stated — lead not stated — Nat Friedman, Daniel Gross, Patrick and John Collison, Arthur Breitman, Andrej Karpathy, Sam Bankman-Fried "among others" (NEW: DEAL-LAB-9014)
Status now
Website live with a London job listing; no 2025–26 news was found in this round.

Labs Round 7 deep profile (30 Oct 2026), as the researchers wrote it; confidence: single_source (all company pages; no 2025–26 news found).

Sources

  1. conjecture.dev/about
  2. conjecture.dev/research/we-are-conjecture-a-new-
  3. conjecture.dev/research/conjecture-2-years
Borough High Street, London (SE1 1JA)Companies House · conjecture.dev/about · find-and-update.company-information.service.gov. · find-and-update.company-information.service.gov. · conjecture.dev/research/we-are-conjecture-a-new-verified
ConsciumCompaniesLondon company that tests and verifies AI agents before deployment and runs a long-term research programme on machine consciousness.Key facts Products: Verifyax (live agent-verification platform), Neuroax (brain-inspired models, "Coming Soon"); research programme "Frontier" on machine consciousness; publications: Principles for Responsible AI Consciousness Research (Butlin and Lappas, 2024) and the book Perspectives on Machine Consciousness (CRC Press, ed. Chace and Lappas); open letter with 173 signatories; partnered with WPP on AI verification (release datelined "London, 18 June"; year not printed in the dateline)
Profile
Founders
Daniel Hulme — founder, CEO and co-founder; Ted Lappas — co-founder; Panos Repoussis — co-founder; Ed Charvet — co-founder; Calum Chace — co-founder
Chief executive
Daniel Hulme
Other leaders
Not stated (only the five co-founders are named)
What it does
Conscium sells a platform that tests and verifies AI agents before and after they are deployed (Verifyax). It is also developing brain-inspired models (Neuroax) and runs a long-term research programme on machine consciousness (Frontier).
Research focus
"Neuroplastic Inference" models; machine consciousness through its Frontier programme, a "long-horizon research programme into machine consciousness".
Models and products
Verifyax — AI agent verification platform (simulation testing, auditing, governance), live with pricing and free trial — date not stated; Neuroax — brain-inspired models, listed "Coming Soon" — date not stated; Frontier — research programme on machine consciousness — date not stated
Customers or partners named
Patrick Butlin (University of Oxford) — co-drafted five principles for responsible AI consciousness research — 2024 — https://conscium.com/frontier
Selected papers
Principles for Responsible AI Consciousness Research (Butlin and Lappas) — PDF on conscium.com, venue not stated — 2024 — https://conscium.com/hubfs/Frontier/Principles-for-Conscious-AI.pdf; Perspectives on Machine Consciousness (book, eds Calum Chace and Ted Lappas) — CRC Press — date not stated — https://conscium.com/frontier
Status now
Trading; Verifyax is offered with pricing and a free trial on its website (read 31 October 2026).
Note
the About page gives founders' other roles — Hulme "Founded Satalia in 2012, sold to WPP in 2021 — now serving as WPP Chief AI Officer"; Repoussis "Leads WPP's AI Research Labs". No Conscium–WPP partnership is stated.

Labs Round 8 deep profile (31 Oct 2026), as the researchers wrote it; confidence: one source.

Sources

  1. conscium.com/
  2. conscium.com/about-us
  3. conscium.com/frontier
Londonconscium.com/frontier · fieldhouseassociates.com/conscium-launches-groun · conscium.com/ · conscium.com/about-usone source
ConvergenceCompaniesLondon start-up that built Proxy, a personal AI agent that learns and carries out multi-step tasks on the web; now part of Salesforce. acquired
Profile
Founded
April 2024 (tech.eu meta description)
Founders
Marvin Purtorab — CEO and co-founder; Andy Toulis — co-founder
Chief executive
Marvin Purtorab — from null
Hq
London (England: London); other_offices: null
What it does
A London start-up that built AI agents able to carry out web-based tasks and cope with pop-ups, errors and page changes. Salesforce bought it in 2025 to strengthen its Agentforce agent product.
Research focus
AI agents for web workflows
Models and products
null (product name not given in fetched sources)
Customers or partners named
Salesforce — acquirer — agreement 15 May 2025, completed 11 June 2025 — https://salesforce.com/news/stories/salesforce-signs-definitive-agreement-to-acquire-convergence-ai/
Funding total as published
12,000,000 USD (pre-seed) — as of 15 May 2025 — https://tech.eu/2025/05/15/salesforce-acquires-uk-ai-startup-convergence-one-year-after-launch/
Valuation reported
null ("Financial details of the deal are not known.")
Rounds found
2024 — pre-seed — 12,000,000 USD — Balderton Capital — Salesforce Ventures and Shopify Ventures named as backers (already DEAL-LAB-032)
Status now
Part of Salesforce since 11 June 2025.

Labs Round 7 deep profile (30 Oct 2026), as the researchers wrote it; confidence: verified.

Sources

  1. salesforce.com/news/stories/salesforce-signs-def
  2. salesforce.com/news/stories/salesforce-signs-def
  3. tech.eu/2025/05/15/salesforce-acquires-uk-ai-sta
Londonsalesforce.com/news/stories/salesforce-signs-def · tech.eu/2025/05/15/salesforce-acquires-uk-ai-sta · tech.eu/2024/09/27/convergence-secures-12m-pre-s · salesforce.com/news/stories/salesforce-signs-defverified
DaltonTxCompaniesLondon company building "Dalton", an AI decision engine that learns from every experiment to help drug discovery teams design small molecules and antibodies.Key facts £4m seed (2025-09-23); Dalton platform launched 2026-06-04; antibody discovery module launched 2026-10-01; Gates Foundation tuberculosis grant 2026-10-15; founded on University of Oxford research (Oxford University Innovation support); named in GOV.UK AI for Science Strategy (2025-11-20); 18 staff named on company site
Profile
Founded
Not stated (emerged from stealth with seed round, 23 Sep 2025)
Founders
Garry Pairaudeau — co-founder and CEO; Anthony Bradley — co-founder and CSO; Adrian Rossall — co-founder; Charlotte Deane — co-founder (Professor of Structural Bioinformatics, University of Oxford)
Chief executive
Garry Pairaudeau (named CEO, 4 Jun 2026)
Chief scientist or research lead
Anthony Bradley — CSO and co-founder
Hq
London, King's Cross (England: London); other_offices: not stated
What it does
DaltonTx sells a drug-discovery software platform, Dalton, that plugs into pharma, biotech and contract-research workflows. It links data, AI models and lab results so scientists can design, make and choose small molecules and antibodies; it calls this an "intelligence backbone" for R&D rather than running its own drug pipeline.
Research focus
Adaptive, continuously learning discovery platform for small molecules and biologics; agentic scientific reasoning; antibody structure analysis with the University of Oxford.
Models and products
Dalton — drug-discovery platform described as a "continuous learning engine" — launched June 2026; Dalton engine antibody extension — antibody design and structure prediction, chat interface — 7 Oct 2026
Customers or partners named
Sygnature Discovery — CRO client/partner on Dalton launch — 4 Jun 2026 — https://pharmaphorum.com/news/meet-dalton-new-ai-powered-drug-discovery-platform; Bonito Biosciences — uses Dalton for CNS oligo therapies — 4 Jun 2026 — https://pharmaphorum.com/news/meet-dalton-new-ai-powered-drug-discovery-platform; University of Oxford — antibody analysis collaboration — 7 Oct 2026 — https://www.bioprocessintl.com/large-molecule-development/daltontx-announces-ai-breakthrough-in-novel-antibody-discovery
Funding total as published
£4m (IQ Capital, tech.eu) / £4.1m (redalpine) — as of 23 Sep 2025 — https://iqcapital.vc/news/daltontx-raises-ps4-million-seed-financing-to-build-the-intelligence-backbone-of-drug-discovery
Rounds found
23 Sep 2025 — seed — £4m / £4.1m — redalpine (per redalpine; IQ Capital names no lead) — IQ Capital Partners, Seedcamp; support from Oxford University Innovation (already DEAL-LAB-4001)
Status now
Trading; launched the Dalton platform in June 2026 and extended it to antibody discovery in October 2026.

Labs Round 8 deep profile (31 Oct 2026), as the researchers wrote it; confidence: verified (seed round: two investor pages plus trade); people and product one source (trade).

Sources

  1. iqcapital.vc/news/daltontx-raises-ps4-million-se
  2. redalpine.com/post/daltontx-the-intelligence-bac
  3. tech.eu/2025/09/23/daltontx-lands-ps4m-to-advanc
  4. pharmaphorum.com/news/meet-dalton-new-ai-powered
  5. bioprocessintl.com/large-molecule-development/da
Londondaltontx.com/news/seed-financing · tech.eu/2025/09/23/daltontx-lands-ps4m-to-advanc · iqcapital.vc/news/daltontx-raises-ps4-million-se · redalpine.com/post/daltontx-the-intelligence-bacverified
DataGemma (RIG and RAG)Chips, models and productsGemma 2 models tuned to answer plain-language questions with public statistics drawn from Google's Data Commons. released (open weights; research tool)Key facts Released 12 Sep 2024; two approaches: Retrieval-Interleaved Generation (RIG) and Retrieval-Augmented Generation (RAG, which feeds Gemini 1.5 Pro); access: open weightsMade by Google DeepMind · 2024-09-12Versions Gemma → Gemma 2 (and CodeGemma, RecurrentGemma, PaliGemma → PaliGemma 2, ShieldGemma → ShieldGemma 2, DataGemma)Londonai.google.dev/gemma/docs/datagemma · ai.google.dev/gemma/docs/releasesone source
DeepFloyd IFChips, models and productsText-to-image model from Stability AI's DeepFloyd research lab, built in three stages and able to write legible text inside images. released (research licence)Key facts Zero-shot FID 6.66 on COCO; T5-XXL-1.1 text encoder; trained on a custom LAION-A set of 1B image-text pairs; stage 1 at 64x64 (400M, 900M, 4.3B parameters), stage 2 to 256x256 (400M, 1.2B), stage 3 to 1024x1024 (700M, not yet released at launch); access: research licenceMade by Stability AI · 28 Apr 2023 (year from page image timestamps)Londonstability.ai/news-updates/deepfloyd-if-text-to-ione source
DeepNash (Stratego agent)Chips, models and productsDeepMind agent that taught itself the hidden-information board game Stratego to human expert level. research (Science paper)Key facts Announced 1 Dec 2022; all-time top-three ranking among human experts on the Gravon Stratego platform; win rate above 97% (often 100%) against top Stratego bots; published in Science; access: paper onlyMade by Google DeepMind · 2022-12-01Londondeepmind.google/blog/mastering-stratego-the-clasone source
DGMR precipitation nowcasting model (with the Met Office)Chips, models and productsRain-forecasting model built with the Met Office that predicts the next 90 minutes of rainfall from radar. research (paper; Nature)Key facts Announced 29 Sep 2021; uses past 20 minutes of radar to predict next 90 minutes; more than 50 Met Office meteorologists rated it first choice in 89% of cases; Nature paper s41586-021-03854-z; access: paper onlyMade by Google DeepMind · 2021-09-29Londondeepmind.google/blog/nowcasting-the-next-hour-ofone source
Dialog-RSN-1Chips, models and productsPolyAI's newest dialogue model, offered by API, which the company says handles conversational turn-taking differently. released (API)Key facts Labelled "Latest model" on PolyAI's technology page; API live with $100 free credit for new accounts for a limited time; access: APIMade by PolyAI · — (exact release date not stated; API live on post of 18 May 2026, page modified 25 Sep 2026)Londonpoly.ai/technology · poly.ai/blog/agentic-dialog-platform-open-to-buione source
Dubbing 2.0 (Synthesia)Chips, models and productsSynthesia's upgraded video dubbing, translating speech into other languages with matching lip movements. live (product)Key facts Released 15 Jul 2026 to all customers; outputs in over 130 languages; Enterprise plans add editing workflow and unlimited dubbingMade by Synthesia · 2026-07-15Londonsynthesia.io/post/introducing-dubbing-2-0 · synthesia.io/blog/category/synthesialab's own pages
Eleven English v1 (eleven_monolingual_v1)Chips, models and productsElevenLabs' first text-to-speech model, trained only on English. deprecatedKey facts Described as the company's "very first model"; trained on an English-only dataset; marked deprecated and due for removal; the 24 Nov 2022 blog describes the beta model (unnamed) as trained on "over 500k hours" of dataMade by ElevenLabs · — (first model; release date not stated)Versions Eleven English v1 → Turbo v2 → Turbo v2.5Londonhelp.elevenlabs.io/hc/en-us/articles/17883183930 · elevenlabs.io/blog/thefirstaithatcanlaughone source
Eleven Turbo v2 (eleven_turbo_v2)Chips, models and productsElevenLabs' first low-latency English speech model, built for real-time use such as phone systems. deprecated (replace with Flash v2)Key facts Announced 9 Nov 2023; about 400 ms latency, over twice as fast as V1 models; output option mulaw 8kHz for VoIP; English only; 0.5 credits per character; now deprecated, replaced by Flash v2 (BIZ-LAB-7024); access: APIMade by ElevenLabs · 2023-11-09Londonelevenlabs.io/blog/turbo-v2-is-here · elevenlabs.io/docs/overview/modelslab's own pages
ElevenLabsCompaniesBuilds AI models that turn text into speech, clone voices and run voice agents, used for audiobooks, dubbing and customer calls.
Profile
Founded
2022 (Series D post; first funding round announced with Eleven v1 in 2022 per tender post)
Founders
Mati Staniszewski — Co-Founder; Piotr Dabkowski — Co-Founder
Chief executive
null (not stated on pages opened)
Other leaders
Ethan Tandowsky — Chief Financial Officer (joined 8 Sep 2026); Ashley Kramer — Chief Revenue Officer (joined 2 Sep 2026); Andrew Reed (Sequoia) — joined board with Series D (Feb 2026)
Hq
London (England: London) — blog post of 8 Jun 2026 says ElevenLabs "expands London HQ"; other_offices: New York, San Francisco, Warsaw, Dublin, Tokyo, Seoul, Singapore, Bengaluru, Sydney, São Paulo, Berlin, Paris, Mexico City (Feb 2026); later launches in Canada (Jul 2026), Spain (Apr 2026), Australia and New Zealand (May 2026); California expansion "creating 173 high-paying jobs" (Jun 2026)
Headcount
over 800 — 30 Sep 2026 — https://elevenlabs.io/blog/tender-22bn; 530 across more than 50 countries — 5 May 2026 — https://elevenlabs.io/blog/500m-arr-and-new-investors (UK headcount not published)
What it does
AI audio company. It builds models for lifelike speech, speech-to-text, music and sound effects, and sells them through ElevenAgents (voice and chat agents for businesses), ElevenCreative (tools for speech, music, sound, image and video) and ElevenAPI (developer access).
Research focus
"AI audio foundation models"; emotional conversational models, dubbing and "audio general intelligence" (Series D post)
Models and products
Eleven Realtime — voice-agent model — 20 Oct 2026; Eleven v4 — "our most emotive model" — 28 Sep 2026; Eleven v4 Turbo — named Sep 2026; ElevenLabs CLI v1 — 24 Aug 2026; Procedures in ElevenAgents — 30 Jun 2026; Ads Engine in ElevenCreative — 22 Jun 2026; Flows Agent in ElevenCreative — 4 Jun 2026; Dubbing v2 — 28 May 2026; Music v2 — 26 May 2026; ElevenMusic — 29 Apr 2026; Eleven v3 generally available — 2 Feb 2026; Scribe v2 — 9 Jan 2026; Scribe v2 Medical — named Sep 2026; Eleven v1 — 2022
Customers or partners named
Universal Music Group — multi-year strategic agreement — 10 Sep 2026 — https://elevenlabs.io/blog; UK Government — voice AI for public services — 8 Jun 2026 — https://elevenlabs.io/blog; NVIDIA (NVentures), Santander — enterprise investors — 5 May 2026 — https://elevenlabs.io/blog/500m-arr-and-new-investors
Funding total as published
781,000,000 USD "across five rounds" — as of 4 Feb 2026 — https://elevenlabs.io/blog/series-d
Valuation reported
11,000,000,000 USD — 4 Feb 2026 — https://elevenlabs.io/blog/series-d; 22,000,000,000 USD — 30 Sep 2026 (employee tender) — https://elevenlabs.io/blog/tender-22bn (company only)
Rounds found
2026-02-04 — Series D — 500,000,000 USD — Sequoia Capital — Andreessen Horowitz, ICONIQ, Lightspeed, Evantic, BOND, BroadLight, NFDG, Valor Capital, AMP Coalition, Smash Capital; 2026-05-05 — Series D third close (amount not stated) — no lead named — BlackRock, Wellington, D.E. Shaw, Schroders, NVentures, Santander, T.Capital, 30+ creative-industry individuals, Robinhood Ventures retail access; separate $100m tender closed; 2026-09-30 — employee tender — 300,000,000 USD — Wellington, T. Rowe Price — BDT & MSD, EQT, GIC, Goldman Sachs, OTPP, Sapphire Ventures (new); Alkeon, a16z, D.E. Shaw, DISRUPTIVE, Evantic, ICONIQ, Lightspeed (existing); 2026-06-18 — "Poland invests in ElevenLabs" (post title only; amount not opened)
Uk government links
Partnership with the UK Government to bring voice AI to public services — 8 Jun 2026; AISI voice-AI security work already on site (BIZ-SAFE-2087)
Status now
Valued at $22bn after a $300m employee tender closed in September 2026; over 800 staff; ARR passed $500m in the first four months of 2026.

Labs Round 7 deep profile (30 Oct 2026), as the researchers wrote it; confidence: single_source (company pages only).

Sources

  1. elevenlabs.io/blog/series-d
  2. elevenlabs.io/blog/series-d
  3. elevenlabs.io/blog/series-d
  4. elevenlabs.io/blog/tender-22bn
  5. elevenlabs.io/blog/tender-22bn
  6. elevenlabs.io/blog/500m-arr-and-new-investors
  7. elevenlabs.io/about
  8. elevenlabs.io/blog
Models and products (21)

Version chainsmusic_v1 → Music v2 (26 May 2026) → Music v2.5 ("Our most advanced music model.")Eleven English v1 → Turbo v2 → Turbo v2.5

  1. Cue (forced alignment model) 2026-10-28 released (API; inside Scribe)A model that lines up each character of a transcript with the exact moment it is spoken in a recording.
  2. Eleven v4 and Eleven v4 Turbo 2026-09-28 released (API and products)ElevenLabs' newest text-to-speech models, producing emotional speech in 90+ languages; the Turbo version is built for live agents.
  3. Dubbing v2 2026-05-28 released (product)An AI dubbing model that translates speech into 90+ languages while keeping the speaker's tone and pacing.
  4. Music v2 (and Music v2.5) 2026-05-26 released (product)ElevenLabs' second music model, with better vocals, instruments and song structure and multi-language lyrics.
  5. Scribe v2 (and Scribe v2 Medical) 2026-01-09 released (API)ElevenLabs' second-generation transcription model for bulk transcripts, subtitles and captions; a medical variant handles clinical audio.
  6. Scribe v2 Realtime 2025-11-11 released (API)A live speech-to-text model for voice agents, meetings and captions.
  7. Eleven Music 2025-08-05 released (product and API)A model that writes full songs from plain-language prompts, with or without vocals, cleared for commercial use.
  8. Voice Design v3 2025-06-25 released (product)A tool that creates a new synthetic voice from a written description and offers three options.
  9. Eleven v3 2025-06-03 released (generally available; previous generation)An expressive text-to-speech model with emotion tags and multi-speaker dialogue in 70+ languages.
  10. Scribe (v1) 2025-02-26 deprecatedElevenLabs' first speech-to-text model, transcribing 99 languages with timings and speaker labels.
  11. Eleven Flash (v2 and v2.5) 2024-12-18 released (API)ElevenLabs' fastest text-to-speech model, built for voice agents that must answer in real time.
  12. Eleven Turbo v2.5 2024-07-19 deprecatedA low-latency text-to-speech model for real-time conversation in 32 languages; now replaced by Flash.
  13. Eleven Turbo v2 (eleven_turbo_v2) 2023-11-09 deprecated (replace with Flash v2)ElevenLabs' first low-latency English speech model, built for real-time use such as phone systems.
  14. Eleven Multilingual v2 2023-08-22 released (API; previous generation)A text-to-speech model that speaks 29 languages, released as ElevenLabs came out of beta.
  15. Eleven Multilingual v1 2023-04-27 supersededElevenLabs' first multilingual text-to-speech model, adding seven languages to its English voice model.
  16. Voice Design (first generative voice model, in Voice Lab) 2023-01-11 superseded (product)ElevenLabs model that creates new synthetic voices from scratch rather than copying a real speaker.
  17. Eleven Realtime v1 (Mini and Max) no date released (API; date not published)Complete speech-to-speech voice-agent models: Mini tuned for speed and natural turn-taking, Max for complex multi-step tasks with tool calls.
  18. SFX v3 (sound effects model) no date limited access (date not published)A model that makes sound effects from text prompts, with better prompt-following and looping.
  19. Eleven English v1 (eleven_monolingual_v1) — (first model; release date not stated) deprecatedElevenLabs' first text-to-speech model, trained only on English.
  20. Speech to Speech voice changer models (English STS v1 and v2; Multilingual STS v2) — (release dates not stated) released (API)ElevenLabs models that turn one person's recorded speech into another voice while keeping the delivery.
  21. Text to Sound Effects v2 (eleven_text_to_sound_v2) — (release date not stated) released (API; default model)ElevenLabs model that makes sound effects from a written description.
Soho, London (W1F 0DQ)Companies House · elevenlabs.io/de/blog/series-d · tech.eu/2026/02/04/elevenlabs-raises-500m-says-b · uktech.news/ai/elevenlabs-valued-at-11bn-after-s · elevenlabs.io/blog/tender-22bnverified
EnformerChips, models and productsModel that predicts how DNA sequence controls gene activity, reading long stretches of DNA at once. released (open model; paper)Key facts Announced 4 Oct 2021; reads 200,000 base pairs of input DNA (more than 5x previous methods); predicts over 5,000 genomic tracks; model and variant predictions openly available; paper in Nature Methods (s41592-021-01252-x); access: open modelMade by Google DeepMind · 2021-10-04Londondeepmind.google/blog/predicting-gene-expression-one source
Faraday (Inherent AI scientist agent)Chips, models and productsInherent's AI "scientist" agent, trained with reinforcement learning to reproduce the findings of published research papers. research releaseKey facts 27B-parameter model (Qwen 3.6 base, per TechCrunch); trained on Replica, a task set built from 100 ML and AI-for-science papers giving 310 tasks; directs OpenAI's GPT-5.5 Codex as a coding tool; company reports it beat Claude Opus 4.8 and GPT-5.5 at replication (no scores published on the page)Made by Inherent · 2026-08-14Londoninherentlabs.ai/research/training-to-replicate · techcrunch.com/2026/08/22/inherent-founded-by-delab's own pages
FermiNet (Fermionic Neural Network)Chips, models and productsNeural network that calculates the quantum behaviour of electrons in molecules from first principles. released (open code; papers)Key facts First published 19 Oct 2020; update 22 Aug 2024 covers a Science paper on excited states; code from the original study and the excited-states work open-sourced on GitHub; access: open codeMade by Google DeepMind · 2020-10-19 (first post; updated 22 Aug 2024 for excited-states Science paper)Londondeepmind.google/blog/ferminet-quantum-physics-an · deepmind.google/blog/page/8/lab's own pages
Frontier AI Research Lab (Thomson Reuters and Imperial College London)Partnerships and contractsA five-year joint lab at Imperial College London, backed by Thomson Reuters, studying the safety, reliability and abilities of frontier AI. announcedKey facts Five-year partnership announced 2 Dec 2025. Research areas: agentic systems, data-centric machine learning, retrieval and grounding, reasoning and planning, evaluation and safety, human-in-the-loop workflows; also the economic and social effects of frontier systems. Plans to host more than a dozen PhD students. Access to Imperial's high-performance computing cluster. UKTN reports the lab will also train large-scale foundation models. Funding amount not disclosed ("guaranteed multi-year funding"). Director: Prof Alessandra Russo; associate directors: Dr Jonathan Richard Schwarz, Prof Felix Steffek.
Profile
Owner or parent
Thomson Reuters and Imperial College London
Founded or opened
announced 2 Dec 2025 (Toronto dateline); five-year partnership
Lab director or lead
Alessandra Russo — Lab Director (per photo notes); Professor in Applied Computational Logic, Imperial Department of Computing; Co-director of the School of Convergence Science, Human and Artificial Intelligence — from Dec 2025
Other leaders
Jonathan Richard Schwarz — Lab Associate Director (Head of AI Research, Thomson Reuters; Visiting Professor at Imperial); Felix Steffek — Lab Associate Director (Professor of Law, University of Cambridge); Mary Ryan — Vice Provost for Research and Enterprise, Imperial
Site and address as published
Imperial College London, "a dedicated facility in Imperial's deep tech community"
Headcount
"more than a dozen PhD students" — 2 Dec 2025 — thomsonreuters.com (single source)
What it does
A five-year research lab at Imperial College London funded by Thomson Reuters. More than a dozen PhD students work with Thomson Reuters scientists and Imperial academics, using Imperial's computing cluster. Topics: AI agents, data-centred machine learning, retrieval and grounding, reasoning and planning, evaluation and safety, and human-in-the-loop work, plus AI's effects on the economy, work and access to justice.
Research groups or focus
agentic systems; data-centric machine learning; retrieval and grounding; reasoning and planning; evaluation and safety; human-in-the-loop workflows; economic impact; future of work; access to justice
Funding or budget as published
"guaranteed multi-year funding", amount not disclosed — 2 Dec 2025 (already DEAL-LAB-6001)
Partners named
NVIDIA — its $250m UK compute commitment names Imperial College London, quoting Russo as "director of Imperial Frontier AI Research Lab" — 13 Oct 2026
Status now
Active: its director was quoted as a named partner in NVIDIA's UK compute commitment on 13 Oct 2026.

Labs Round 8 deep profile (31 Oct 2026), as the researchers wrote it; confidence: verified (Thomson Reuters release and NVIDIA release; UKTN already on record).

Sources

  1. thomsonreuters.com/en/press-releases/2025/decemb
  2. nvidianews.nvidia.com/news/nvidia-commits-250-mi
Models and products (1)
  1. Snowdon (Frontier AI Research Lab model) 2026-08 released (used as Thomson base model)A re-aligned version of open Qwen models built by the Thomson Reuters–Imperial lab and used as the base for Thomson.
Londonthomsonreuters.com/en/press-releases/2025/decemb · uktech.news/ai/thomson-reuters-backed-frontier-a · thomsonreuters.com/content/dam/ewp-m/documents/t · nvidianews.nvidia.com/news/nvidia-commits-250-miverified
Gemini 2.0 Flash Thinking (experimental)Chips, models and productsFirst Gemini model released in a "thinking" mode that reasons step by step before answering. superseded (public preview)Key facts Public preview 19 Dec 2024; updated preview gemini-2.0-flash-thinking-exp-01-21 released 21 Jan 2025; access: API (preview); later Gemini 2.5 thinking models are listed as BIZ-LAB-7130Made by Google DeepMind · 2024-12-19Londonai.google.dev/gemini-api/docs/changelogone source
Gemini DiffusionChips, models and productsExperimental Google DeepMind language model that writes text by diffusion, refining whole passages at once rather than word by word. research (experimental demo)Key facts Sampling speed stated as 1,479 tokens per second excluding overhead (overhead 0.84 seconds), averaged across reported evaluations; access: experimental demo with sign-up for updates; DiffusionGemma (BIZ-LAB-7161) is stated to build on its researchMade by Google DeepMind · — (release date not stated on the model page)Londondeepmind.google/models/gemini-diffusion/one source
Gemma 2 (2B, 9B, 27B; plus Gemma 2 JPN)Chips, models and productsSecond generation of Google DeepMind's small open-weight language models, released in three sizes. released (open weights)Key facts 9B and 27B released 27 Jun 2024; 2B released 31 Jul 2024; Gemma 2 JPN (2B) released 3 Oct 2024; 27B trained on 13 trillion tokens, 9B on 8 trillion, 2B on 2 trillion (model card); trained on TPUv5p; access: open weightsMade by Google DeepMind · 2024-06-27Versions Gemma → Gemma 2 (and CodeGemma, RecurrentGemma, PaliGemma → PaliGemma 2, ShieldGemma → ShieldGemma 2, DataGemma)Londonai.google.dev/gemma/docs/releases · ai.google.dev/gemma/docs/core/model_card_2lab's own pages
Genie (first version, "Generative Interactive Environments")Chips, models and productsFirst Genie world model: generates interactive, controllable environments, learned without labels from internet videos. research (paper only)Key facts arXiv 2402.15391 submitted 23 Feb 2024; 11B parameters; trained unsupervised on unlabelled internet videos; access: paper only (successors Genie 2, 3 and 4 already listed)Made by Google DeepMind · 2024-02-23Versions Genie → Genie 2 → 3 → 4Londonarxiv.org/abs/2402.15391one source
Google DeepMindCompaniesBuilds general-purpose AI models and research systems for Google and for science, including the Gemini models and the Genie world-simulation models. merged
Profile
Founded
2010 (DeepMind); Google DeepMind formed 20 April 2023 when DeepMind and the Brain team from Google Research merged
Founders
Demis Hassabis — co-founder and CEO (other co-founders not named on pages opened)
Chief executive
Demis Hassabis — CEO of the combined Google DeepMind from 20 April 2023
Chief scientist or research lead
Koray Kavukcuoglu — led the new Scientific Board, with the research leadership team reporting to him (April 2023; current title not checked)
Other leaders
Eli Collins — VP of Product (April 2023); Zoubin Ghahramani — research leadership team, reporting to Koray Kavukcuoglu (April 2023); Lila Ibrahim, Pushmeet Kohli, Owen Larter — named co-authors with Demis Hassabis of the 10 December 2025 UK partnership post (roles not stated on that page)
Hq
null (careers page lists London first but does not call it headquarters); other_offices: Bay Area, Bangalore, Cambridge (US), Montreal, New York City, Paris, Tokyo, Toronto, Zurich (careers page, read 30 Oct 2026)
What it does
Alphabet's AI research unit, formed in 2023 by merging DeepMind with Google's Brain team. It builds the Gemini family of models, generative media models (Veo for video, Lyria for music, Nano Banana for images, Genie for interactive worlds), Gemini Robotics models, the open Gemma models, and science systems such as AlphaFold, AlphaGenome, AlphaEvolve and WeatherNext.
Research focus
"build AI responsibly to benefit humanity"; areas listed on its site: frontier models (Gemini), generative media, world models, robotics, games and agents (SIMA), AI for science (AlphaFold, AlphaGenome, WeatherNext, AlphaEarth, AlphaEvolve, Co-Scientist), frontier safety
Models and products
Gemini 4 Carbon — "Our next frontier model" — October 2026; Gemini 4 Argon (and Gemini 4 Argon for Robotics) — October 2026; Genie 4 — world simulation model — October 2026; EmbeddingGemma 2 — open multimodal embedding model — October 2026; Gemini Robotics 2 — robot "intelligence layer" — 30 July 2026; Gemini Robotics On-Device 2 — runs locally on robots — 30 July 2026; Gemini Robotics ER 2 — embodied reasoning model — July 2026; Gemini Robotics-ER 1.6 — April 2026; Gemini 3.8 Flash — "most intelligent workhorse model yet for coding and agents" — date not shown; Gemma 4 — open models — date not shown; Gemini Omni 1.1 Flash — date not shown; Lyria 3.5 — music model — date not shown; Nano Banana 2 Lite — image model — date not shown; Gemini Robotics 1.5 — September 2025; Gemini Robotics — March 2025
Customers or partners named
UK government (i.AI), Google Cloud, Faculty, and Barnet, Dorset and Camden planning authorities — AI planning-officer prototype — 16 June 2026 — https://deepmind.google/blog/unlocking-uk-house-building-with-ai-accelerated-planning/; Apptronik, Boston Dynamics, Agile Robots — thanked for support on Gemini Robotics 2 — 30 July 2026 — https://deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots/; Singapore Government — fine-tuned model for scam detection — October 2026 — https://deepmind.google/
Papers and awards
AlphaFold described as "Nobel-winning" (Isomorphic Labs homepage; year not given on page) — https://isomorphiclabs.com/?amp=&amp=
Funding total as published
null (unit of Alphabet)
Rounds found
none (merger already held as DEAL-LAB-001)
Uk government links
Partnership announced 10 December 2025 — UK scientists get priority access to AlphaEvolve, AlphaGenome, AI co-scientist and WeatherNext; Google DeepMind's first automated laboratory, focused on materials science, to open in the UK in 2026; deeper work with the AI Security Institute; AI cyber-defence work (Big Sleep, CodeMender); education research on Gemini and England's national curriculum. Planning: i.AI's Gemini-based "Extract" made available to every council in England in June 2026; a national planning tool planned for all councils from 2027. Already on site as BIZ-BIG-2009, BIZ-SAFE-2085 and BIZ-SVC-030 / DEAL-SVC-013.
Status now
Operating as Alphabet's AI unit; its homepage lists Gemini 4 Carbon, Gemini 4 Argon and Genie 4 among October 2026 news.

Labs Round 7 deep profile (30 Oct 2026), as the researchers wrote it; confidence: single_source (company pages only).

Sources

  1. deepmind.google/about/
  2. deepmind.google/blog/announcing-google-deepmind/
  3. deepmind.google/blog/strengthening-our-partnersh
  4. deepmind.google/blog/unlocking-uk-house-building
  5. deepmind.google/blog/gemini-robotics-2-brings-wh
  6. deepmind.google/
  7. deepmind.google/careers/
Models and products (140)

Version chainsLyria → Lyria 2Gemini Robotics (Mar 2025) → On-Device (Jun 2025) → 1.5 (Sep 2025) → ER 1.6 (Apr 2026) → Gemini Robotics 2 / ER 2 / On-Device 2 (Jul 2026) → Gemini 4 Argon for Robotics (Oct 2026)Gemma → Gemma 2 (and CodeGemma, RecurrentGemma, PaliGemma → PaliGemma 2, ShieldGemma → ShieldGemma 2, DataGemma)Genie 1 (2D worlds) → Genie 2 (Dec 2024) → Genie 3 (Aug 2025) → Genie 4 (Oct 2026)Veo 2 (Dec 2024) → Veo 3 → Veo 3.1 (Oct 2025) → Veo 3.1 Lite (card 8 Apr 2026)AlphaGeometry → AlphaGeometry 2Genie → Genie 2 → 3 → 4AlphaCode → AlphaCode 2AlphaGo → AlphaZero → MuZeroAlphaGo Fan → AlphaGo Lee → AlphaGo Master → AlphaGo Zero

  1. Gemini 4 Carbon 2026-10-28 released (announced)Google DeepMind's newest frontier Gemini model, announced two days before this ledger's cut-off.
  2. Genie 4 2026-10-27 research preview (trusted testers; waitlist)Google DeepMind's newest world model, generating interactive worlds in real time with sound, physics and characters steered by text.
  3. Gemini 4 Argon 2026-10-13 released (apps, API, AI Studio, Antigravity)The first Gemini 4 model, aimed at software engineering, law, finance and defensive security work.
  4. Gemini 4 Argon for Robotics 2026-10-13 releasedA robotics version of Gemini 4 Argon.
  5. SynthID Bio 2026-09-30 proof of conceptA proof of concept for watermarking AI-designed proteins without changing how they work.
  6. Gemini 3.8 audio models (Live, Live Extended Thinking, TTS, Live Avatar) 2026-09-15 releasedGemini's September 2026 real-time voice, text-to-speech and talking-avatar models.
  7. AlphaGenome Atlas 2026-09-08 released (free portal for academic research; API)A free catalogue predicting the molecular effects of about 9 billion possible single-letter changes in the human genome.
  8. Gemini 3.8 Flash 2026-09-02 releasedGoogle DeepMind's current Flash model, aimed at complex agent tasks at scale.
  9. Gemini 3.7 Flash 2026-08-13 releasedAn August 2026 Gemini Flash model.
  10. Lyria 3.5 2026-07-29 released (Google Flow Music; Gemini)Google DeepMind's newest music model, with better musicality, lyrics and vocals.
  11. Gemini Robotics 2 (with Robotics-ER 2 and On-Device 2) 2026-07-28 releasedGoogle DeepMind's second-generation robot models, adding whole-body control, video understanding and multi-robot teamwork.
  12. Gemini 3.6 Flash and Gemini 3.5 Flash-Lite 2026-07-21 releasedMid-2026 Gemini Flash and Flash-Lite models for faster, cheaper tasks.
  13. Gemini 3.5 Flash Cyber (and 3.8 Flash Cyber) 2026-07-17 releasedA lightweight Gemini model for finding and patching software security flaws.
  14. Nano Banana 2 Lite (Gemini 3.1 Flash-Lite Image) and Nano Banana 2.1 2026-06-30 releasedLighter and updated versions of the Nano Banana 2 image model.
  15. DiffusionGemma 2026-06-10 released (open weights)An open Gemma model that generates text by diffusion, which the lab says is four times faster.
  16. Gemini 3.5 Live Translate and Gemini 3.5 Transcribe 2026-06-09 releasedGemini speech models for near real-time spoken translation and for speech-to-text transcription.
  17. AlphaFold Protein Structure Database 2026-05-19 activeA free public database of AI-predicted protein structures, made by Google DeepMind's AlphaFold and run with EMBL-EBI.
  18. Gemini Omni (Omni, Omni Flash, Omni 1.1 Flash) 2026-05-17 releasedGoogle DeepMind's model for creating content "from anything", starting with video.
  19. Gemini 3.5 Flash 2026-05-15 released (apps, API, Antigravity)The first Gemini 3.5 model, built for complex, multi-step agent tasks.
  20. Co-Scientist 2026-05-12 releasedA Gemini-based multi-agent AI partner built to help researchers speed up scientific work.
  21. AI co-clinician 2026-04-30 researchGoogle DeepMind research towards an AI that works alongside clinicians in patient care.
  22. Decoupled DiLoCo 2026-04-22 researchA Google DeepMind method for resilient training of AI across distributed computers.
  23. Gemini Robotics-ER 1.6 2026-04-13 releasedA robotics reasoning model with better spatial and multi-camera understanding.
  24. Gemma 4 (E2B, E4B, 12B, 26B, 31B) 2026-04-02 released (open weights)Google DeepMind's fourth generation of open-weight models, built for reasoning and agent workflows.
  25. Gemini 3.1 Flash Live and Gemini 3.1 Flash TTS 2026-03-26 releasedGemini voice models: Flash Live for real-time spoken conversation and Flash TTS for directed, expressive speech.
  26. Lyria 3 Pro 2026-03-25 releasedA Lyria music model that writes longer tracks with song structure.
  27. Gemini 3.1 Flash-Lite 2026-03-03 releasedThe fastest, cheapest Gemini 3 series model at its release.
  28. Nano Banana 2 (Gemini 3.1 Flash Image) 2026-02-26 releasedA fast Gemini image generation model with Pro-level features.
  29. Gemini 3.1 Pro 2026-02-19 released (API and apps)Google DeepMind's February 2026 Pro model for complex reasoning and coding tasks.
  30. Lyria 3 2026-02-18 released (Gemini app, Vertex AI, AI Studio, YouTube, API)Google DeepMind's music generation model that turns a text prompt into a song; it made music creation available in the Gemini app.
  31. Gemini 3 Deep Think 2026-02-12 released (updated reasoning mode)Gemini's most specialised reasoning mode, updated for science, research and engineering problems.
  32. Project Genie 2026-01-29 released (prototype for Google AI Ultra subscribers in the US)An experimental app in which users create and explore AI-generated interactive worlds.
  33. D4RT 2026-01-16 researchA model for reconstructing and tracking scenes in four dimensions (3D plus time), claimed to be much faster than earlier methods.
  34. Gemini 3 Flash 2025-12-17 releasedA fast, lower-cost Gemini 3 model.
  35. Gemma Scope 2 2025-12-16 released (open tools)Open interpretability tools that let researchers inspect the inner workings of every Gemma 3 model.
  36. FunctionGemma and TranslateGemma 2025-12 released (open weights)Specialised open Gemma models: one for calling software functions, one for translation.
  37. Nano Banana Pro (Gemini 3 Pro Image) 2025-11-20 released (Gemini app and API)Google DeepMind's Gemini 3 Pro image generation and editing model.
  38. Gemini 3 Pro 2025-11-18 released (superseded by 3.1 Pro)The first Gemini 3 model, launched in November 2025 for consumers and developers.
  39. SIMA 2 2025-11-13 limited research previewA Gemini-powered game agent that follows instructions, reasons about goals, talks with users and improves itself in 3D worlds.
  40. Google Antigravity (and Antigravity 2.0) 2025-11 released (product)A Google product, launched November 2025, through which Gemini models and DeepMind tools (such as an AlphaGenome skill) are offered; version 2.0 followed in May 2026.
  41. WeatherNext 2 2025-11 releasedGoogle DeepMind's November 2025 weather forecasting model.
  42. CodeMender 2025-10-06 research (not public; human-reviewed patches)An AI agent that finds and fixes security flaws in software and rewrites code to make it safer.
  43. Gemini 2.5 Computer Use 2025-10 released (model card)A Gemini model that operates computer interfaces such as web pages by clicking and typing.
  44. Veo 3 and Veo 3.1 (incl. Veo 3.1 Lite) 2025-10 released (Gemini app, Flow, Gemini API, AI Studio, Vids)Google DeepMind's leading video model, which makes 8-second clips at 1080p or 4K.
  45. Gemini Robotics 1.5 and Gemini Robotics-ER 1.5 2025-09-25 released (ER 1.5 via Gemini API; 1.5 to select partners)A robot action model that reasons before acting, paired with a planning model that can call tools such as Google Search.
  46. VaultGemma 2025-09 released (open weights)A Gemma language model trained with differential privacy, a privacy-protecting training method.
  47. EmbeddingGemma (and EmbeddingGemma 2) 2025-09 released (open weights)Small open Gemma models that turn text (and, in version 2, other media) into numbers for search and retrieval.
  48. Genie 3 2025-08-05 limited research previewA world model that builds explorable, interactive virtual worlds from a text description in real time.
  49. Gemini 2.5 Flash Image 2025-08 releasedA Gemini image generation and editing model; the models page brands Gemini Image as "Nano Banana".
  50. AlphaEarth Foundations 2025-07-30 released (Satellite Embedding dataset in Google Earth Engine)An AI model that combines satellite and other Earth data into one map-like representation for tracking land and environment.
  51. Aeneas 2025-07-23 released (free tool; open code and data)An AI model that helps historians date, place and restore ancient Latin inscriptions.
  52. MedGemma (incl. MedGemma 1.5 4B) 2025-07 released (open weights)Open Gemma-based models for building health AI applications.
  53. T5Gemma (and T5Gemma 2) 2025-07 released (open weights)A collection of Gemma models with an encoder-decoder design.
  54. AlphaGenome 2025-06-25 released (API preview for non-commercial research)An AI model that predicts how single DNA changes affect the processes that switch genes on and off.
  55. Gemini Robotics On-Device 2025-06-24 released (SDK for trusted testers)A robotics model for two-armed robots that runs on the robot itself, without a network connection, and can be fine-tuned.
  56. Gemma 3n 2025-06 released (open weights)An open Gemma model variant introduced in June 2025 with a developer guide.
  57. Imagen 4 2025-05-20 releasedGoogle DeepMind's Imagen 4 image generation model.
  58. AlphaEvolve 2025-05-14 released (early access for academics)Google DeepMind's coding agent that uses Gemini models plus automatic checkers to discover and improve algorithms.
  59. SynthID Detector 2025-05 released (portal)A portal to help identify AI-generated content marked with SynthID.
  60. Lyria 2 (and Music AI Sandbox update) 2025-04-24 released (waitlist, US)Second Lyria music model, offered to musicians and producers in the US through Google's Music AI Sandbox.
  61. DolphinGemma 2025-04 research releaseA Gemma-based model used to help decode dolphin communication.
  62. Gemini Robotics and Gemini Robotics-ER 2025-03-12 released (trusted testers; Apptronik partnership)Gemini-based models that let robots act on what they see and are told; the ER version adds spatial reasoning for robot builders.
  63. ShieldGemma 2 2025-03-10 released (open weights)Open model that checks images for harmful content against set safety policies.
  64. Gemini 2.5 (Pro, Flash, Flash-Lite, Deep Think) 2025-03 released (superseded by Gemini 3)Gemini's "thinking" model generation of 2025; an advanced Deep Think version reached gold-medal standard at the International Mathematical Olympiad.
  65. Gemma 3 (incl. Gemma 3 270M) 2025-03 released (open weights)The third generation of Google's open-weight Gemma models; a tiny 270-million-parameter version followed in August 2025.
  66. Gemini 2.0 Flash Thinking (experimental) 2024-12-19 superseded (public preview)First Gemini model released in a "thinking" mode that reasons step by step before answering.
  67. PaliGemma 2 (and PaliGemma 2 mix) 2024-12-05 released (open weights)Second version of the open image-and-text Gemma model, in three sizes.
  68. Genie 2 2024-12-04 research onlyA world model that turns one image into a playable 3D world a person or AI agent can move through.
  69. GenCast 2024-12-04 released (open code and weights)An AI weather model that gives a range of possible forecasts up to 15 days ahead, including extreme-weather risk.
  70. Gemini 2.0 (Flash, Flash-Lite) 2024-12 released (superseded)The Gemini generation Google DeepMind introduced for the "agentic era", with Flash and Flash-Lite models for developers.
  71. Veo 2 2024-12 released (superseded)Google DeepMind's second video generation model, released with Imagen 3 in December 2024.
  72. Imagen 3 2024-12 released (superseded)Google DeepMind's Imagen 3 image generation model, announced with Veo 2.
  73. FACTS Grounding and FACTS Benchmark Suite 2024-12 released (benchmarks)Google DeepMind benchmarks that test how factually accurate language models are.
  74. AlphaQubit 2024-11 research releaseAn AI system aimed at quantum error correction (per its post address).
  75. Multi-speaker dialogue generation (behind NotebookLM Audio Overviews) 2024-10-30 live (inside Google products)Google DeepMind speech technology that writes out two-person podcast-style conversations as audio, used in NotebookLM Audio Overviews.
  76. AlphaChip 2024-09-26 released (open checkpoint and weights)A reinforcement-learning method that lays out computer chips, used for Google's TPU designs.
  77. Veo (first version) in YouTube Dream Screen 2024-09-18 (YouTube integration announced; May 2024 launch post on blog.google not opened) superseded (product integration)Google DeepMind's first Veo video model, brought to YouTube Shorts to make short background clips from a chosen image.
  78. DataGemma (RIG and RAG) 2024-09-12 released (open weights; research tool)Gemma 2 models tuned to answer plain-language questions with public statistics drawn from Google's Data Commons.
  79. ALOHA Unleashed and DemoStart 2024-09-12 research (paper only)Two robot-learning methods: one teaches two-armed robots complex tasks, the other trains dexterous hands in simulation.
  80. AlphaProteo 2024-09-05 released (paper only)An AI system that designs new proteins that bind tightly to chosen target molecules.
  81. Gemma Scope 2024-07-31 released (open sparse autoencoders)Open tools that let safety researchers look inside Gemma 2 models to see how they process information.
  82. ShieldGemma 2024-07-31 released (open weights)Open Gemma 2 models that check text going into and out of AI systems for harmful content.
  83. AlphaProof 2024-07-25 research (paper; competition result)Reinforcement-learning system that proves maths statements in the formal language Lean; solved three IMO 2024 problems.
  84. AlphaGeometry 2 2024-07-25 research (competition result)Improved version of AlphaGeometry, pairing a Gemini-based language model with a symbolic engine to solve olympiad geometry.
  85. Gemma 2 (2B, 9B, 27B; plus Gemma 2 JPN) 2024-06-27 released (open weights)Second generation of Google DeepMind's small open-weight language models, released in three sizes.
  86. V2A (video-to-audio research) 2024-06-17 research (not released)Research system that generates soundtracks, effects and dialogue to match the action in a video.
  87. PaliGemma 2024-05-14 released (open weights)Small open model that reads images and text together, for tasks such as captioning and answering questions about pictures.
  88. AlphaFold 3 and AlphaFold Server 2024-05-08 released (free server; code and weights for academic use)The third AlphaFold, which predicts the structures and interactions of proteins and other molecules, offered through a free web server.
  89. Med-Gemini 2024-04-29 research (paper only)Family of Gemini models specialised for medicine, tested on medical exam and imaging benchmarks.
  90. CodeGemma 2024-04-09 released (open weights)Small open Gemma models trained to complete, write and chat about computer code.
  91. RecurrentGemma (2B and 9B) 2024-04-09 released (open weights)Open Gemma language model built on a different design (Griffin) that mixes recurrence with local attention to run more efficiently.
  92. TacticAI 2024-03-19 released (paper; Liverpool FC collaboration)An AI assistant that advises football coaches on corner kicks by predicting outcomes and suggesting player positions.
  93. SIMA (Scalable Instructable Multiworld Agent) 2024-03-13 research only (technical report)An AI agent that follows spoken-style instructions to carry out tasks in 3D video games using keyboard and mouse.
  94. Genie (first version, "Generative Interactive Environments") 2024-02-23 research (paper only)First Genie world model: generates interactive, controllable environments, learned without labels from internet videos.
  95. Gemini 1.5 2024-02 released (superseded)The second Gemini generation; a May 2024 update added a faster model and a longer context window.
  96. Gemma (first generation) 2024-02 released (open weights)Google's first family of lightweight open-weight models, built from the technology behind Gemini.
  97. AlphaGeometry 2024-01-17 released (paper; open code and model)A system that pairs a language model with a logic engine to prove Olympiad geometry problems.
  98. FunSearch 2023-12-14 released (paper)A system in which a language model writes programs and an automatic checker scores them, used to find new maths results.
  99. Gemini 1.0 2023-12 released (superseded)Google DeepMind's first Gemini model family, described at launch as its largest and most capable AI model.
  100. GNoME (Graph Networks for Materials Exploration) 2023-11-29 released (paper and data)A deep-learning tool that predicts which new crystal materials are stable.
  101. Lyria (first version) and Dream Track 2023-11-16 released (limited: YouTube experiment)Google DeepMind's first Lyria music-generation model, launched with a YouTube Shorts experiment that made songs in artists' voices.
  102. GraphCast 2023-11-14 released (open code)An AI weather model that makes 10-day global forecasts in under a minute.
  103. Open X-Embodiment dataset and RT-X models (RT-1-X, RT-2-X) 2023-10-03 released (open dataset; models)Pooled robot-training dataset from many labs and the robot models trained on it, which work across different robot types.
  104. AlphaMissense 2023-09-19 released (open catalogue CC BY 4.0; open code)An AI model that estimates whether single-letter changes in human proteins are likely to cause disease.
  105. SynthID 2023-08-29 released (in Google products; text watermarking open-source planned May 2024)An invisible watermark that marks AI-generated images, then text, video and audio, so they can be identified later.
  106. RT-2 (Robotic Transformer 2) 2023-07-28 released (paper)A robot-control model that learns from web and robot data and turns camera images and instructions into robot actions.
  107. RoboCat 2023-06-20 research (paper only)Self-improving robot agent that learns new tasks on different real robot arms from a few hundred demonstrations.
  108. AlphaDev 2023-06-07 released (paper; algorithms open-sourced into LLVM libc++)An AI system that found faster sorting and hashing routines by building programs one low-level instruction at a time.
  109. Imagen 2 2023 (exact day not stated on source) superseded (released)Second version of Google's Imagen text-to-image model, with an added model for judging image aesthetics.
  110. AlphaCode 2 2023 (reported in DeepMind review dated 22 Dec 2023) research (technical report)Second AlphaCode system for competitive programming, solving more contest problems than the first.
  111. DeepNash (Stratego agent) 2022-12-01 research (Science paper)DeepMind agent that taught itself the hidden-information board game Stratego to human expert level.
  112. AlphaTensor 2022-10-05 released (paper and code)An AI system that searches for faster ways to do matrix multiplication and checks they are correct.
  113. Sparrow (dialogue agent) 2022-09-22 research (paper only)Research chatbot trained with human feedback and rules to give evidence-backed answers and avoid unsafe replies.
  114. Gato 2022-05-12 released (paper)One neural network with one set of weights that plays games, captions images, chats and controls a robot arm.
  115. MuJoCo (open-sourced physics simulator) 2022-05 released (open source)A physics simulator for robotics research that DeepMind opened up in 2021 and open-sourced in 2022.
  116. Flamingo 2022-04-28 released (paper)A single model that reads images, video and text together and learns new tasks from a few examples.
  117. Chinchilla 2022-04-12 released (paper)A 70-billion-parameter language model that showed training on more data can beat simply adding parameters.
  118. Ithaca 2022-03 released (paper)An AI model for ancient Greek inscriptions; the predecessor of Aeneas.
  119. AlphaCode 2022-02-02 released (paper and dataset)A system that writes computer programs for programming contests by generating many candidates and filtering them.
  120. Gopher 2021-12-08 released (paper)A 280-billion-parameter language model that DeepMind used to study scale, risks and retrieval.
  121. RETRO (Retrieval-Enhanced Transformer) 2021-12-08 research (paper only)Language model that looks up passages in a very large text database while it writes, so a smaller model can match much larger ones.
  122. Enformer 2021-10-04 released (open model; paper)Model that predicts how DNA sequence controls gene activity, reading long stretches of DNA at once.
  123. DGMR precipitation nowcasting model (with the Met Office) 2021-09-29 research (paper; Nature)Rain-forecasting model built with the Met Office that predicts the next 90 minutes of rainfall from radar.
  124. Perceiver and Perceiver IO 2021-08-03 released (open code; papers)General neural-network designs that take in many kinds of data (images, sound, text) and produce many kinds of output.
  125. XLand open-ended learning agents 2021-07-27 research (paper only)DeepMind agents trained across hundreds of thousands of generated 3D games that learned general game-playing behaviour.
  126. MuZero 2020-12-23 released (paper)A game-playing system that learns its own model of a game and plans with it, without being told the rules.
  127. AlphaFold2 2020-11-30 released (open code)The second AlphaFold, which predicts protein shapes close to lab-measured accuracy; organisers called it a solution to protein folding.
  128. FermiNet (Fermionic Neural Network) 2020-10-19 (first post; updated 22 Aug 2024 for excited-states Science paper) released (open code; papers)Neural network that calculates the quantum behaviour of electrons in molecules from first principles.
  129. Agent57 2020-03 released (paper)A DeepMind game-playing agent reported to outperform the human benchmark on Atari games.
  130. AlphaStar 2019-01-24 released (research result)An AI program that plays the full real-time strategy game StarCraft II and beat top professional players.
  131. AlphaZero 2018-12-06 released (paper)One learning system that teaches itself chess, shogi and Go from the rules alone, by playing against itself.
  132. AlphaFold (first version, CASP13) 2018-12-02 released (paper)The first AlphaFold, which predicts a protein's 3D shape from its sequence; it ranked first at the CASP13 benchmark in 2018.
  133. Streams (DeepMind Health app) 2017-11 released (NHS hospital app)A mobile app for NHS hospital clinicians; DeepMind published a post explaining why it did not use AI.
  134. AlphaGo Zero 2017-10-18 released (paper)A Go-playing system that taught itself the game from random moves by playing against itself, with no human game records.
  135. WaveNet (in Google Assistant) 2017-10 released (in product)DeepMind's raw-audio speech generation model (first published 2016), launched in Google Assistant in October 2017.
  136. WeatherNext 3 — (no date on page) releasedGoogle DeepMind's AI global weather model, which produces a new forecast every hour and feeds Google Search, Maps and Gemini.
  137. Magenta RealTime and Lyria RealTime no date released (Magenta RealTime open; dates not on page)Real-time interactive music generation models; Magenta RealTime is the open Lyria model.
  138. Gemini Embedding no date released (date not on page)A Gemini model that turns text and other media into numbers for search and retrieval.
  139. Gemini Diffusion — (release date not stated on the model page) research (experimental demo)Experimental Google DeepMind language model that writes text by diffusion, refining whole passages at once rather than word by word.
  140. Project Astra — (date not stated on the project page) research prototype (trusted testers)Google DeepMind's prototype universal assistant that sees and hears through a phone camera or prototype glasses and talks back in real time.
London (EC2M 4AG)Companies House · deepmind.google/blog/announcing-google-deepmind/ · deepmind.google/about · find-and-update.company-information.service.gov. · find-and-update.company-information.service.gov.verified
HaiperCompaniesLondon company that built a generative AI model making short video clips from text prompts. dormant
Profile
Founded
2021
Founders
Yishu Miao — co-founder and CEO (former DeepMind; led TikTok's London ML team); Ziyu Wang — co-founder and CTO (former DeepMind and Google Brain)
Chief executive
Yishu Miao — from null
Hq
London (England: London); other_offices: null
What it does
A London company that built a visual foundation model for generating short video clips from text, image and video prompts, aimed at publishers, studios and individuals. Its Haiper 1.5 release (July 2024) made 8-second clips with a 1080p upscaler.
Research focus
video generation; visual foundation models
Models and products
Haiper 1.5 — text/image/video-to-video, 8-second clips, 1080p upscaler — 16 July 2024
Funding total as published
19,200,000 USD (£15.1m) — as of 6 March 2024 — https://www.businesscloud.co.uk/news/video-generative-ai-platform-raises-10-8m/ (trade)
Rounds found
2024-03-06 — seed — 13,800,000 USD (£10.8m) — Octopus Ventures (per BusinessCloud; deal file says "not named") — others not named (already DEAL-LAB-031 / DEAL-LAB-2030)
Status now
null — no news after July 2024 found in this round.

Labs Round 7 deep profile (30 Oct 2026), as the researchers wrote it; confidence: single_source.

Sources

  1. uktech.news/ai/haiper-ai-video-generation-202403
  2. businesscloud.co.uk/news/video-generative-ai-pla
  3. venturebeat.com/ai/exclusive-meet-haiper-1-5-the
Models and products (3)

Version chainsHaiper Video 1.5 → 2.x

  1. Haiper Video 1.5 — (release date not stated on docs) released (API and app)Haiper's earlier video model that makes short clips from a text prompt or a still image.
  2. Haiper Video 2.0 (2.x) — (release date not stated on docs; launch post on haiper.ai not opened) released (API and app)Haiper's second-generation video model for text-to-video and image-to-video clips.
  3. Haiper Image 1.5 and Haiper Image 2.x — (release dates not stated) released (API)Haiper's text-to-image models, offered alongside its video models.
Northdown Street, London (N1 9BN)Companies House · uktech.news/ai/haiper-ai-video-generation-202403 · find-and-update.company-information.service.gov. · uktech.news/tag/haiper · find-and-update.company-information.service.gov.verified
Haiper Image 1.5 and Haiper Image 2.xChips, models and productsHaiper's text-to-image models, offered alongside its video models. released (API)Key facts Image 2.x priced $0.04 (720p), $0.08 (1080p), $0.12 (4K) per image; Image 1.5 priced $0.02 (720p) and $0.04 (1080p); text-to-image endpoint resolutions 720, 1080 and 2160; access: APIMade by Haiper · — (release dates not stated)Londondocs.haiper.ai/api-reference/endpoint/2-0-text-t · docs.haiper.ai/api-reference/api-pricingone source
Haiper Video 1.5Chips, models and productsHaiper's earlier video model that makes short clips from a text prompt or a still image. released (API and app)Key facts Clip durations 2, 4 or 8 seconds; aspect ratios 21:9 to 1:1; camera moves (pan, tilt, zoom); API price $0.05 per second at 720p and $0.033 at 540p; extend and 1080p upscale functions; access: API and appMade by Haiper · — (release date not stated on docs)Versions Haiper Video 1.5 → 2.xLondondocs.haiper.ai/api-reference/endpoint/1-5-text-i · docs.haiper.ai/api-reference/api-pricingone source
Haiper Video 2.0 (2.x)Chips, models and productsHaiper's second-generation video model for text-to-video and image-to-video clips. released (API and app)Key facts API "gen2" endpoints; text-to-video clips of 4 or 6 seconds (default 6); modes standard, smooth and enhanced; API price $0.05 per second at 720p, $0.033 at 540p; access: API and appMade by Haiper · — (release date not stated on docs; launch post on haiper.ai not opened)Londondocs.haiper.ai/introduction · docs.haiper.ai/api-reference/api-pricingone source
Imagen 2Chips, models and productsSecond version of Google's Imagen text-to-image model, with an added model for judging image aesthetics. superseded (released)Key facts Released 2023; improved outputs via a specialised image aesthetics model; successors Imagen 3 and 4 already listed (BIZ-LAB-7164, 7165)Made by Google DeepMind · 2023 (exact day not stated on source)Londondeepmind.google/blog/2023-a-year-of-groundbreakione source
InherentCompaniesLondon research lab building Faraday, an AI system meant to let scientists and self-improving AI work together on research.
Profile
Legal name
null (site footer reads "© 2026 Inherent Laboratories"; the site says it is incorporated as a Public Benefit Corporation; Companies House not checked)
Founders
Tantum Collins — Co-Founder & CEO; Edward Hughes — Co-Founder & Chief Scientist; Louis Kirsch — Co-Founder; Kaloyan Aleksiev — Co-Founder & CTO
Chief executive
Tantum Collins — from null
Chief scientist or research lead
Edward Hughes — Co-Founder & Chief Scientist
Other leaders
Matt Clifford — adviser (tech.eu)
Hq
London (England: London); other_offices: null
What it does
A London AI research lab building "general-purpose inventive AI" to speed up scientific research. Its system, Faraday, is meant to let people and self-improving AI work together on hard scientific problems. It says it applies recursive self-improvement to its models and to how it runs research.
Research focus
recursive self-improvement; AI Scientist agents; open-ended research; human–machine teaming; autonomous labs
Models and products
Faraday — AI system for scientific discovery — announced May 2026; reported in August 2026 to run on Qwen 3.6 with 27 billion parameters (trade, single)
Papers and awards
"Training AI Scientists to Replicate Research" — 14 August 2026 (inherentlabs.ai/research)
Funding total as published
50,000,000 USD — as of 29 May 2026 — https://tech.eu/2026/05/29/london-based-ai-lab-inherent-emerges-from-stealth-with-50m-raise/
Rounds found
2026-05-29 — seed — 50,000,000 USD — Index Ventures and Radical Ventures (co-leads) — none named (already DEAL-LAB-027 / DEAL-LAB-2009)
Uk government links
null (tech.eu describes adviser Matt Clifford as a former government AI adviser)
Status now
Active London research lab that left stealth in May 2026 and published its first research post in August 2026.

Labs Round 7 deep profile (30 Oct 2026), as the researchers wrote it; confidence: verified (founders, roles, round: investor page plus trade); Faraday model details single_source.

Sources

  1. radical.vc/portfolio/inherent/
  2. tech.eu/2026/05/29/london-based-ai-lab-inherent-
  3. inherentlabs.ai/
  4. inherentlabs.ai/research/
  5. newsbytesapp.com/news/science/deepmind-alumni-s-
Models and products (1)
  1. Faraday (Inherent AI scientist agent) 2026-08-14 research releaseInherent's AI "scientist" agent, trained with reinforcement learning to reproduce the findings of published research papers.
LondonCompanies House · tech.eu/2026/05/29/london-based-ai-lab-inherent- · uktech.news/ai/london-based-ai-lab-inherent-emer · inherentlabs.ai/ · find-and-update.company-information.service.gov.verified
Isomorphic LabsCompaniesBuilds AI models that design medicines, which it uses in its own drug programmes and in work with drug companies.
Profile
Founders
Sir Demis Hassabis — Founder and CEO
Chief executive
Sir Demis Hassabis — from null
Chief scientist or research lead
Miles Congreve — Chief Scientific Officer; Sarah Skerratt — Chief Research Officer
Other leaders
Max Jaderberg — President (from January 2026); Sarah Korman — Chief Business and Legal Officer; Beverley Mallon — Chief People & Experience Officer; Dr Ben Wolf — Chief Medical Officer; Colin Murdoch — described as "Isomorphic Labs' first President" (article title; dates not opened). Scientific Advisory Board: Jennifer Doudna, Sir David MacMillan, Sir Paul Nurse, Sir Menelas Pangalos, Venki Ramakrishnan
Hq
London (England: London); other_offices: Cambridge, Massachusetts; Lausanne, Switzerland
What it does
AI drug discovery company that builds on the AlphaFold system. It develops its own AI drug design engine (IsoDDE), runs an in-house therapeutic pipeline it is taking towards the clinic, and works with drug makers including Novartis, Lilly and Johnson & Johnson.
Research focus
"predictive and generative AI models" for drug design; AI drug design engine (IsoDDE)
Models and products
IsoDDE — AI drug design engine; funds from the May 2026 round go to its development and deployment — 12 May 2026
Customers or partners named
Novartis, Lilly, Johnson & Johnson — named partnerships — 12 May 2026 — https://www.isomorphiclabs.com/press/isomorphic-labs-funding; Eli Lilly — multi-target research collaboration, $45m upfront — 7 Jan 2024 — https://www.isomorphiclabs.com/articles/isomorphic-labs-kicks-off-2024-with-two-pharmaceutical-collaborations; Novartis — research collaboration, $37.5m upfront — 7 Jan 2024 — same url; Virtual Biology Initiative — joined (homepage item shown dated 18.02.25) — https://isomorphiclabs.com/?amp=&amp=
Papers and awards
builds on the "Nobel-winning AlphaFold system" (homepage; year not given)
Funding total as published
null (no total published on pages opened)
Rounds found
2025-03-31 — first external round — 600,000,000 USD — Thrive Capital — GV; Alphabet (follow-on); 2026-05-12 — Series B — 2,100,000,000 USD — Thrive Capital — Alphabet, GV (existing); MGX, Temasek, CapitalG, UK Sovereign AI Fund (new). UKTN reports the round as £1.47bn.
Uk government links
UK Sovereign AI Fund invested in the May 2026 Series B
Status now
Raised a $2.1bn Series B in May 2026 to develop IsoDDE and move its pipeline towards the clinic; led by Demis Hassabis with Max Jaderberg as President.

Labs Round 7 deep profile (30 Oct 2026), as the researchers wrote it; confidence: verified (May 2026 round: company + UKTN); other fields single_source.

Sources

  1. isomorphiclabs.com/press/isomorphic-labs-funding
  2. isomorphiclabs.com/press/isomorphic-labs-funding
  3. isomorphiclabs.com/articles/isomorphic-labs-anno
  4. isomorphiclabs.com/articles/isomorphic-labs-anno
  5. isomorphiclabs.com/our-team
  6. isomorphiclabs.com/articles/in-conversation-with
  7. isomorphiclabs.com/articles/isomorphic-labs-kick
  8. isomorphiclabs.com/?amp=&amp=
Models and products (2)
  1. Isomorphic Labs Drug Design Engine (IsoDDE) 2026-02-10 internal use (technical report)Isomorphic Labs' combined computer system that predicts molecular structures, binding strength and binding pockets to design drugs.
  2. Dataverse (Isomorphic Labs) no date internal use (date not published)Isomorphic Labs' curated collection of life-science data that feeds its research environment.
London (EC2M 4AG)Companies House · isomorphiclabs.com/press/isomorphic-labs-funding · isomorphiclabs.com/articles/isomorphic-labs-anno · find-and-update.company-information.service.gov. · tech.eu/2026/05/13/isomorphic-labs-lands-2-1bn-iverified
Juno-N-Coder-25B (Locai Labs)Chips, models and productsLocai Labs' compressed open-source coding model, cut down from NVIDIA Nemotron 3.5 Lightning to run on its Locai 1 office server. releasedKey facts 25.57B total parameters (19% fewer than the 31.56B base); experts pruned by 25%; quantised to NVFP4; listed as 25B-A3B; first Juno model shipped on the Locai 1 appliance (press release 11 Aug 2026)Made by Locai Labs · 2026-07Londonlocailabs.com/research · locailabs.com/news/press-release-locai-labs-launone source
Jupiter-N (Locai Labs)Chips, models and productsLocai Labs' 120-billion-parameter model: NVIDIA's Nemotron 3 Super re-trained for British context; it powers Locai's GB1 assistant. releasedKey facts 120B parameters; post-train of NVIDIA Nemotron 3 Super; technical report arXiv 2604.17429; a sister 8B model, Jupiter-G, applies the same recipe to Google Gemma-4 (locailabs.com/research). Civo calls Jupiter the first model of the Mercury Series (14 Apr 2026)Made by Locai Labs · 2026-04Londonlocailabs.com/research · civo.com/blog/uks-first-sovereign-llms-locai-lablab's own pages
Latent LabsCompaniesBuilds AI models and an AI agent that design new proteins, such as antibodies, for drug and biotechnology companies.
Profile
Founders
Dr Simon Kohl — Founder & CEO (previously co-lead of DeepMind's protein design team)
Chief executive
Simon Kohl — from null
Other leaders
Annette Obika-Mbatha — Head of Operations; Stefan Oschmann — strategic advisory board (Dec 2025)
Hq
London (England: London); other_offices: San Francisco (wet lab)
Headcount
null (team page lists 13 people; not stated as headcount)
What it does
Builds generative AI models that design new proteins from scratch — binders, antibodies and macrocyclic peptides — for chosen disease targets, offered through a no-code web platform and an API.
Research focus
generative protein design; de novo binders, antibodies (VHH and scFv) and macrocycles
Models and products
Latent-X2 — designs drug-like antibodies and macrocycles; low ex vivo immunogenicity in a ten-donor panel — 16 Dec 2025; Latent-X — protein binder design; wet-lab hit rates 91–100% for macrocycles and 10–64% for mini-binders across 7 targets — 22 Jul 2025
Customers or partners named
null (Latent-X2 opened to "selected partners", none named)
Papers and awards
null (technical reports appear in search results only — see Gaps)
Funding total as published
50,000,000 USD — as of 13 Feb 2025 — https://www.latentlabs.com/press-release/latent-labs-secures-50m-in-funding/
Rounds found
2025-02-13 — Series A (within $50m total) — 40,000,000 USD — Radical Ventures, Sofinnova Partners (co-leads) — Flying Fish, Isomer; 8VC, Kindred Capital, Pillar VC (existing); angels Jeff Dean, Aidan Gomez, Mati Staniszewski
Status now
Released Latent-X2 for antibody design in December 2025; no funding after February 2025 found.

Labs Round 7 deep profile (30 Oct 2026), as the researchers wrote it; confidence: single_source.

Sources

  1. latentlabs.com/press-release/latent-labs-secures
  2. latentlabs.com/press-release/latent-labs-secures
  3. latentlabs.com/press-release/latent-labs-announc
  4. latentlabs.com/press-release/introducing-latent-
  5. latentlabs.com/team/
Models and products (3)

Version chainsLatent-X (Jul 2025) → Latent-X2 (Dec 2025) → Latent-Y agent (Mar 2026)

  1. Latent-Y 2026-03-23 released (selected partners; Research Tier for researchers)An AI agent that designs antibodies from written goals, running Latent-X2 inside an automated workflow.
  2. Latent-X2 2025-12-16 early access (web and API; commercial use by partnership)Latent Labs' second model, which designs antibodies and ring-shaped peptides with properties suited to drug development.
  3. Latent-X 2025-07-22 released (no-code web platform, free tier)Latent Labs' first public AI model, which designs small protein binders and ring-shaped peptides ready for lab testing.
White Collar Factory, Old Street Yard, London (EC1Y 8AF)Companies House · latentlabs.com/press-release/latent-labs-secures · tech.eu/2025/02/13/ai-pioneer-dr-simon-kohls-lat · latentlabs.com/press-release/latent-y/ · latentlabs.com/press-release/verified
Lumen Sovereign (Cosine)Chips, models and productsCosine's planned UK-trained frontier model, co-designed with 13 British banks, defence firms and others, to run inside customers' own systems. in training (target late 2026; commercial launch Q1 2027)Key facts Target: 1T-parameter intermediate checkpoint with 49.2B active parameters, 256k-context sparse mixture-of-experts; 500,000 GH200 GPU-hour budget on Isambard-AI via the UK Sovereign AI programme; about one-third of compute for software engineering, two-thirds for areas such as cyber testing and KYC/AML; coalition of 13 (BAE Systems, Babcock, BT, Era4, HSBC, Leonardo, Lloyds Banking Group, LSEG, NatWest Group, PwC, Telefónica Tech UK&I, Thales UK, The Alan Turing Institute) announced 8 Jun 2026Made by Cosine · 2026-06-08Londoncosine.sh/lumen-sovereign · cosine.sh/blog/building-lumen-sovereign-uk-induslab's own pages
Lyria (first version) and Dream TrackChips, models and productsGoogle DeepMind's first Lyria music-generation model, launched with a YouTube Shorts experiment that made songs in artists' voices. released (limited: YouTube experiment)Key facts Announced 16 Nov 2023; Dream Track experiment with nine artists for a limited set of creators; all Lyria output watermarked with SynthID; access: limited product testMade by Google DeepMind · 2023-11-16Versions Lyria → Lyria 2Londondeepmind.google/blog/transforming-the-future-of- · deepmind.google/blog/2023-a-year-of-groundbreakilab's own pages
Lyria 2 (and Music AI Sandbox update)Chips, models and productsSecond Lyria music model, offered to musicians and producers in the US through Google's Music AI Sandbox. released (waitlist, US)Key facts Announced 24 Apr 2025 within Music AI Sandbox; access widened to more US musicians, producers and songwriters via waitlist; output watermarked with SynthID; Lyria RealTime named in same post (already listed as BIZ-LAB-7172)Made by Google DeepMind · 2025-04-24Versions Lyria → Lyria 2Londondeepmind.google/blog/music-ai-sandbox-now-with-n · deepmind.google/models/lyria/one source
ManticCompaniesLondon start-up building an AI system that forecasts world events.
Profile
Founded
2024 (Reuters)
Founders
Toby Shevlane — co-founder and CEO (former Google DeepMind research scientist); Ben Day — co-founder (machine learning PhD, Cambridge)
Chief executive
Toby Shevlane — from null
Hq
London (England: London); other_offices: null
What it does
A London start-up building AI for "judgmental forecasting": predicting political, economic and business events such as supply-chain disruption or geopolitical shocks. In the summer 2026 Metaculus Cup its system beat every human entrant and all but one bot. Hedge funds and trading firms are reported as interested buyers.
Research focus
AI forecasting of world events
Models and products
Mantic forecasting system — 2025–26
Customers or partners named
null (company declined to name customers, September 2026)
Papers and awards
Metaculus Cup, summer 2026 — outperformed every human contestant and all but one bot (bot named laertes)
Funding total as published
null (rounds: $4m pre-seed August 2025; $25m seed September 2026)
Valuation reported
null ("undisclosed valuation")
Rounds found
2025-08-27 — pre-seed — 4,000,000 USD — Episode 1 — DRW and angel investors (already DEAL-LAB-035 / DEAL-LAB-2029); 2026-09-18 — seed — 25,000,000 USD — Radical Ventures — M12 (Microsoft), Thinking Machines Lab, Balderton Capital, others (NEW: DEAL-LAB-9003)
Status now
Active London start-up; raised a $25m seed in September 2026.

Labs Round 7 deep profile (30 Oct 2026), as the researchers wrote it; confidence: single_source (no primary source opened; mantic.com blocked).

Sources

  1. whbl.com/?p=1008351
  2. thestar.com.my/tech/tech-news/2026/09/18/ai-star
  3. tech.eu/2025/08/27/mantic-which-claims-to-foreca
Models and products (1)
  1. Mantic forecasting system (Scan, Forecast, Monitor) — (launch date not stated) live (product, by demo)Mantic's AI that gives probabilities, with reasoning and sources, for future world events and updates them as news changes.
Kensal Road, London (W10 5BN)Companies House · tech.eu/2025/08/27/mantic-which-claims-to-foreca · find-and-update.company-information.service.gov. · find-and-update.company-information.service.gov. · uktech.news/ai/ai-startup-that-predicts-world-evone source
Mantic forecasting system (Scan, Forecast, Monitor)Chips, models and productsMantic's AI that gives probabilities, with reasoning and sources, for future world events and updates them as news changes. live (product, by demo)Key facts Product modules Scan, Forecast and Monitor; site states it beat all 676 human forecasters in the Metaculus Cup (Summer 2025 label); Polymarket backtest calibration error 0.022 vs 0.058 for Polymarket on markets with $1k+ traded; blog (5 Oct 2026) reports a 45.12% forecast for Lula against a 45.16% result; access: product by demoMade by Mantic · — (launch date not stated)Londonmantic.com/ · mantic.com/newslab's own pages
Meta AI research, London hubData centres and compute sitesOne of Meta's four EMEA AI research hubs; London and Paris researchers open-sourced the Implicitron 3D modelling framework.
Profile
Owner or parent
Meta (Fundamental AI Research, FAIR)
Lab director or lead
Not stated for London. Naila Murray — head of FAIR EMEA (regional role) — Jun 2023
Other leaders (historical; current roles not checked)
Sebastian Riedel — Facebook AI Research Manager, dual UCL affiliation — Feb 2021; Tim Rocktäschel and Edward Grefenstette — FAIR London researchers with dual UCL affiliation — Feb 2021; Andrea Vedaldi — joined the London office (also Oxford associate professor) — Jul 2018
Site and address as published
London (no address)
What it does
Meta's Fundamental AI Research team in London, one of four EMEA hubs with Paris, Tel Aviv and Zurich. Pages opened show London work on computer vision and 3D (Implicitron, open-sourced with Paris in 2022), on language (the Bloomsbury AI team joined in 2018), and a UK PhD programme with UCL from 2021. No 2025–26 page about the London team was found.
Research groups or focus (shared by the four EMEA hubs)
self-supervised learning; reinforcement learning; speech and audio; computer vision; natural language modelling; responsible AI; machine learning theory; model efficiency; AR/VR
Notable outputs
Implicitron — modular framework in PyTorch3D that learns 3D objects or scenes from sparse images, open-sourced by London and Paris researchers — Aug 2022
Partners named
University College London — FAIR UK PhD programme, four students in year one — Feb 2021; University of Oxford — through Andrea Vedaldi — Jul 2018
Uk government links
DSIT's sector plan update (10 Jun 2026) names Meta among AI firms in London's King's Cross cluster
Status now
Not stated: no 2025–26 page about the London research hub was opened.

Labs Round 8 deep profile (31 Oct 2026), as the researchers wrote it; confidence: one source.

Sources

  1. about.fb.com/news/2023/06/a-spotlight-on-the-fou
  2. ai.meta.com/blog/facebook-ai-research-expands-wi
  3. ai.meta.com/blog/fair-unveils-uk-phd-program-in-
Londonabout.fb.com/news/2023/06/a-spotlight-on-the-fou · ai.meta.com/blog/facebook-ai-research-expands-wi · ai.meta.com/blog/fair-unveils-uk-phd-program-in-one source
Metis ReasoningCompaniesNew company developing an AI reasoning system meant to be easier to check and to need far less computing power than today's large models.
Profile
Legal name
null (site footer "© 2026 Metis Reasoning")
Founded
null (public launch 20 October 2026)
Founders
Thore Graepel — co-founder (AlphaGo co-creator; left Google DeepMind summer 2026); Piers Ingram — co-founder; Michael Warren — co-founder (robotics, ex-Boeing); David Berman — co-founder (Head of AI, Cambridge Consultants)
Hq
London (England: London); other_offices: Singapore; Los Angeles (as listed in the white paper, per The Next Web)
Headcount
12 — October 2026 — https://tech.eu/2026/10/20/former-google-deepmind-top-researcher-launches-london-compute-efficient-ai-company/ (trade; The Next Web repeats tech.eu)
What it does
A London lab working on AI reasoning that it says needs much less compute than frontier large language models. It wants systems that keep an auditable record of what they know and how sure they are, for high-stakes work in robotics, science and engineering. It has published a white paper but no model or test results.
Research focus
"generalizable machine reasoning"; embodied AI, science and engineering; approach drawn from AlphaGo/AlphaZero self-play and Jaakko Hintikka's inquirer-versus-nature game (white paper, per The Next Web)
Models and products
The Age of Reasoning — white paper — October 2026; no model released
Papers and awards
"The Age of Reasoning" white paper — October 2026
Compute
says its design is "much much more compute-efficient" than frontier models — October 2026
Funding total as published
null (reported to be raising tens of millions of dollars; the company would not confirm the amount)
Uk government links
David Berman described as a UK government and NATO AI adviser (tech.eu) — October 2026
Status now
Launched publicly in London on 20 October 2026; no product or results released.

Labs Round 7 deep profile (30 Oct 2026), as the researchers wrote it; confidence: single_source (founders and headcount from trade only; the company site names no people).

Sources

  1. tech.eu/2026/10/20/former-google-deepmind-top-re
  2. thenextweb.com/news/metis-reasoning-thore-graepe
  3. metisreasoning.ai/
LondonCompanies House · tech.eu/2026/10/20/former-google-deepmind-top-re · metisreasoning.ai/ · find-and-update.company-information.service.gov. · find-and-update.company-information.service.gov.verified
Multi-speaker dialogue generation (behind NotebookLM Audio Overviews)Chips, models and productsGoogle DeepMind speech technology that writes out two-person podcast-style conversations as audio, used in NotebookLM Audio Overviews. live (inside Google products)Key facts Post 30 Oct 2024; produces 2 minutes of dialogue in under 3 seconds on a single TPU v5e chip, over 40 times faster than real time; underpins NotebookLM Audio Overviews and Illuminate; model not named; access: inside Google productsMade by Google DeepMind · 2024-10-30Londondeepmind.google/blog/pushing-the-frontiers-of-auone source
Open X-Embodiment dataset and RT-X models (RT-1-X, RT-2-X)Chips, models and productsPooled robot-training dataset from many labs and the robot models trained on it, which work across different robot types. released (open dataset; models)Key facts Announced 3 Oct 2023; data from 22 robot types; more than 500 skills, 150,000 tasks, more than 1 million episodes; partners from 33 academic labs; RT-1-X outperformed the original model by 50% on average; access: open datasetMade by Google DeepMind · 2023-10-03Londondeepmind.google/blog/scaling-up-learning-across- · deepmind.google/blog/2023-a-year-of-groundbreakilab's own pages
OpenAI London research hubData centres and compute sitesOpenAI's London research team, planned as its largest research hub outside San Francisco, working on frontier model development.
Profile
Owner or parent
OpenAI
Founded or opened
first international office, London, announced 28 Jun 2023; named a major research hub on 26 Feb 2026
Lab director or lead
Phoebe Thacker — global head of data research programmes and London site lead — Apr 2026
Other leaders
Mark Chen — chief research officer, OpenAI (quoted Feb 2026); Diane Yoon — VP of People (2023)
Site and address as published
new office at Regent Quarter, King's Cross (Jahn Court and the Brassworks Building), 88,500 sq ft, room for up to 544 staff, opening 2027
Headcount
Not stated (544 is capacity, not staff)
What it does
OpenAI's London base, its first office outside the US (2023). In February 2026 OpenAI said London would become its biggest research hub outside San Francisco, with London researchers owning "key components" of frontier model development, including quality, reliability, alignment and safety. Its Training team hires London researchers to build OpenAI's main language models.
Research groups or focus
Training team ("responsible for producing the large language models that power our research"); frontier model development: quality, reliability, alignment and safety
Notable outputs
Not stated (none credited to London)
Partners named
Endurance Land — developer of the King's Cross office — Apr 2026
Uk government links
OpenAI–UK MoU signed by Sam Altman and Peter Kyle on 21 Jul 2025 (record BIZ-BIG-003); Technology Secretary Liz Kendall welcomed the research hub (Feb 2026); £5.6m to AISI's Alignment Project (Feb 2026); £100m in compute and model credits for the AI Biosecurity Lab (Oct 2026)
Status now
Active and growing; its permanent King's Cross office is due in 2027.

Labs Round 8 deep profile (31 Oct 2026), as the researchers wrote it; confidence: verified (OpenAI pages and UKTN).

Sources

  1. uktech.news/ai/openai-names-london-as-its-next-m
  2. uktech.news/ai/openai-secures-first-permanent-lo
  3. openai.com/careers/researcher-training-london-lo
  4. openai.com/index/introducing-openai-london/
  5. uktech.news/ai/openai-expands-uk-footprint-and-b
London, King's Cross (new Regent Quarter office, about 88,500 sq ft, expected 2027)uktech.news/ai/openai-names-london-as-its-next-m · uktech.news/ai/openai-secures-first-permanent-lo · openai.com/careers/researcher-training-london-lo · openai.com/index/introducing-openai-london/one source
Orbital (Applied Computing energy foundation model)Chips, models and productsApplied Computing's AI for oil refineries and petrochemical plants, combining plant sensor data, physics and a chemical-engineering language model. live (product)Key facts Three parts: time-series model on raw plant data, physics-based model that extracts equations from manuals and literature, and a language model trained on chemical-engineering knowledge; company figures (28 May 2025): outperforms benchmarked software peers by about 90% on key metrics, average 75% cost cut vs cloud AI, payback under 10 weeks, uses 100% of facility data vs 8%; case study (28 Aug 2024) on a simulation of Reliance's Jamnagar refinery found $119K annual hydrotreater savings; INSITE 3.0 "powered by Orbital" with KBR shown 24 Mar 2026; runs natively on AWS; not related to Orbital Industries (BIZ-LAB-024) or Orbital (BIZ-SU-LEG-011)Made by Applied Computing · — (flagship product named in release of 28 May 2025; first public date not stated)Londonappliedcomputing.com/press-release-28052025 · appliedcomputing.com/more/partnershipslab's own pages
Orbital Industries (Orbital Materials Ltd)CompaniesLondon company that uses its own atom-level AI simulation engine, Orb, to design materials and build cooling and other hardware for AI data centres. renamed
Profile
Legal name
null on pages opened (site holds "Orbital Materials Ltd"); traded as Orbital Materials until May 2026
Founded
2022
Founders
Jonathan Godwin — co-founder and CEO (formerly DeepMind); James Gin-Pollock — co-founder and CTO; Daniel Miodovnik — co-founder and COO
Chief executive
Jonathan Godwin — from 2022
Chief scientist or research lead
James Gin-Pollock — CTO
Hq
London (England: London) per Datacentre News; other_offices: San Francisco
Headcount
about 50 — 12 Jun 2026 — (trade) https://datacentrenews.uk/story/orbital-raises-usd-50-million-to-scale-data-centres
What it does
Uses its AI materials platform to design data-centre hardware: a non-toxic, PFAS-free two-phase cooling fluid for dense GPU racks and modular data centres built off-site (branded Orbital IT). It also publishes Orb, an open-source model that simulates atom-level behaviour.
Research focus
AI for materials discovery; atomic-scale simulation; data-centre cooling and infrastructure
Models and products
Orb (v3) — open-source interatomic simulation model, Apache 2.0; company claims stable runs of up to 100,000 atoms on one GPU — reported May 2026; PFAS-free dielectric two-phase cooling fluid — May 2026; modular data-centre system (Orbital IT), deployable "in as little as six months" (company claim) — May 2026
Customers or partners named
AWS — strategic agreement (Dec 2024, under the old name); Orb on SageMaker JumpStart and AWS Marketplace — https://thenextweb.com/news/orbital-industries-50m-series-b-data-center-cooling; SLB — manufacturing partner — reported 12 Jun 2026 — https://datacentrenews.uk/story/orbital-raises-usd-50-million-to-scale-data-centres
Rounds found
2026-05-28 — Series B — 50,000,000 USD — Plural (partner Ian Hogarth) — NVentures, Radical Ventures, Compound, Fly Ventures (already held as DEAL-LAB-036)
Status now
Renamed from Orbital Materials and raised a $50m Series B in May 2026 to build AI-designed cooling and modular data centres; about 50 staff.

Labs Round 7 deep profile (30 Oct 2026), as the researchers wrote it; confidence: single_source (two trade sources; company site could not be opened); sources are listed in the record panel.

Models and products (6)

Version chainsOrb (2024) → OrbMol (27 Aug 2025) → Orb-v3 (4 Sep 2025) → OrbMol-v2 (2 Jun 2026)

  1. OrbMol-v2 2026-06-02 released (blog)Orbital's second molecular model, which it says outperforms the most-cited method in chemistry.
  2. Orb-v3 2025-09-04 released (open source, Apache 2.0)A family of Orb models that simulate materials atom by atom at large scale, trading precision against computing cost.
  3. OrbMol 2025-08-27 released (open source, Apache 2.0)An Orb model extended to molecules, including metal compounds, biological molecules and battery electrolytes.
  4. Orb (universal interatomic potential) 2024-09-02 released (open source)Orbital's open-source AI model that simulates how atoms in advanced materials behave, faster than older methods.
  5. LINUS (Orbital foundation model) no date internal (date not published)Orbital's in-house foundation model for materials, on which Orb is built.
  6. CurieOS no date released (deployed by BASF ECMS; date not published)Orbital's software for AI-driven materials discovery, used in chemistry, hardware and engineering.
London (WC1N 2EB)Companies House · tech.eu/2026/05/28/orbital-industries-secures-50 · find-and-update.company-information.service.gov. · find-and-update.company-information.service.gov. · find-and-update.company-information.service.gov.verified
Owl (PolyAI speech recognition model)Chips, models and productsPolyAI's own speech-to-text model, built to hear callers accurately over phone lines with noise and accents. live (inside PolyAI platform)Key facts Announced 3 Jun 2025; word error rate 0.122, stated as most accurate of the models PolyAI tested; PolyAI says about 70% of its voice agents' errors come from speech recognition; used in PolyAI Agent Studio; access: within PolyAI platform (no standalone release stated)Made by PolyAI · 2025-06-03Londonpoly.ai/blog/owl-speech-recognition-model-polyai · poly.ai/technologylab's own pages
PaliGemmaChips, models and productsSmall open model that reads images and text together, for tasks such as captioning and answering questions about pictures. released (open weights)Key facts Initial release 14 May 2024; 3B parameters; inspired by PaLI-3 (arXiv 2310.09199), uses SigLIP vision encoder and Gemma language model; access: open weightsMade by Google DeepMind · 2024-05-14Versions Gemma → Gemma 2 (and CodeGemma, RecurrentGemma, PaliGemma → PaliGemma 2, ShieldGemma → ShieldGemma 2, DataGemma)Londonai.google.dev/gemma/docs/paligemma · ai.google.dev/gemma/docs/releaseslab's own pages
PaliGemma 2 (and PaliGemma 2 mix)Chips, models and productsSecond version of the open image-and-text Gemma model, in three sizes. released (open weights)Key facts Released 5 Dec 2024 in 3B, 10B and 28B sizes; PaliGemma 2 mix released 19 Feb 2025 in 3B, 10B and 28B; access: open weightsMade by Google DeepMind · 2024-12-05Versions Gemma → Gemma 2 (and CodeGemma, RecurrentGemma, PaliGemma → PaliGemma 2, ShieldGemma → ShieldGemma 2, DataGemma)Londonai.google.dev/gemma/docs/releases · ai.google.dev/gemma/docs/paligemmalab's own pages
Perceiver and Perceiver IOChips, models and productsGeneral neural-network designs that take in many kinds of data (images, sound, text) and produce many kinds of output. released (open code; papers)Key facts Blog 3 Aug 2021; papers arXiv 2103.03206 (Perceiver) and 2107.14795 (Perceiver IO); code open-sourced in deepmind-research GitHub repository; access: open codeMade by Google DeepMind · 2021-08-03Londondeepmind.google/blog/building-architectures-thatone source
Personal Avatars (Synthesia digital twins)Chips, models and productsSynthesia feature that builds a talking digital copy of a real person from a photo, webcam or studio footage. live (paid plans)Key facts Photo avatars built on Express-2 technology, ready in minutes after consent; video avatars ready in 24 hours; Studio avatar is a paid add-on from green-screen footage; consent video must be recorded live; photo avatars on paid plans onlyMade by Synthesia · — (launch date not stated; docs page modified 22 Oct 2026)Londondocs.synthesia.io/docs/personal-avatarsone source
PhysicsXCompaniesBuilds AI models that predict how physical products such as aircraft parts, cars and yachts will behave, so engineers can design and test them faster.
Profile
Founders
Jacomo Corbo — Co-Founder & CEO; Robin Tuluie — Founder and Chairman
Chief executive
Jacomo Corbo — from null
Chief scientist or research lead
Dael Williamson — Chief Technology and Platform Officer (from 1 Oct 2026)
Other leaders
Chris Wigley — COO (appointed 12 Nov 2025)
Hq
London (England: London) — Victoria House, 1 Leonard Circus, EC2A 4DQ; other_offices: New York; Singapore (Asia-Pacific hub, Oct 2026); growing presence in San Francisco Bay Area
Headcount
more than 300 — 8 Jun 2026 — https://www.physicsx.ai/newsroom/physicsx-announces-300m-series-c-to-accelerate-physics-ai-for-industrial-engineering; 20 in Singapore, target over 100 within three years — 8 Oct 2026 (UK headcount not published)
What it does
Builds AI models that predict physical behaviour — airflow, heat, stress — so engineers can test and optimise designs far faster than with conventional simulation. Sold as an AI-native engineering platform to aerospace and defence, automotive, semiconductor, energy and materials companies.
Research focus
"Large Physics Models" (larger pre-trained physics AI models, in development); generative optimisation; neural operators; agentic engineering
Models and products
PX-Car-Aero-v1 — automotive aerodynamics model — 7 Aug 2026; Large Physics Models — in development — 8 Jun 2026; AI-native engineering platform — ongoing
Customers or partners named
Fortescue — expanded partnership — 27 Oct 2026 — https://www.physicsx.ai/newsroom; GB1 (Britain's America's Cup challenge) — yacht design — 13 Mar 2026 and 21 Oct 2026 — same url; Seatrium — physics AI for offshore and marine design — 8 Oct 2026 — https://www.physicsx.ai/newsroom/physicsx-launches-asia-pacific-hub-in-singapore-to-scale-ai-powered-industrial-engineering-across-the-region; A*STAR — five-year partnership, joint S$40m commitment — 8 Oct 2026 — same url; Siemens — data centre power infrastructure (with NVIDIA) — 17 Mar 2026 — https://www.physicsx.ai/newsroom; CoreWeave — compute partnership — 11 Mar 2026 — same url; Deutsche Telekom — multi-year partnership — 18 Nov 2025 — same url; Microsoft — selected for Agentic Launchpad — 18 Dec 2025 — same url; Siemens — deep physics simulation — 4 Dec 2024 — https://www.physicsx.ai/newsroom/siemens-physicsx-collaborate-to-build-ai-based-deep-physics-simulation
Compute
CoreWeave partnership for high-performance physics AI — 11 Mar 2026; live on the European Industrial AI Cloud — 4 Feb 2026 (newsroom titles)
Valuation reported
about 2,400,000,000 USD — 8 Jun 2026 — company only
Rounds found
2026-06-08 — Series C — 300,000,000 USD — Temasek — M&G Investments, Intrepid Growth Partners (new); Applied Materials, Atomico, General Catalyst, July Fund, NGP, NVIDIA, Radius, Siemens (existing); 2025-11-19 — Series B extension (amount not opened)
Uk government links
British Business Bank in the Series C (held as DEAL-INV-004); no government contract found
Status now
Raised a $300m Series C at about $2.4bn in June 2026 and opened a Singapore hub in October 2026.

Labs Round 7 deep profile (30 Oct 2026), as the researchers wrote it; confidence: single_source.

Sources

  1. physicsx.ai/newsroom/physicsx-announces-300m-ser
  2. physicsx.ai/newsroom/physicsx-announces-300m-ser
  3. physicsx.ai/newsroom/physicsx-announces-300m-ser
  4. physicsx.ai/newsroom
  5. physicsx.ai/newsroom/physicsx-launches-asia-paci
  6. physicsx.ai/newsroom/siemens-physicsx-collaborat
Models and products (4)
  1. PX-Car-Aero-v1 2026-08-07 announced (article)A PhysicsX physics-AI model for car aerodynamics.
  2. PhysicsX Large Physics Models on CoreWeave (private, domain-specific models) 2026-03-11 released (enterprise)A PhysicsX service that lets industrial customers train private physics AI models on their own data on CoreWeave's cloud.
  3. LGM-Aero (Large Geometry Model for aerospace) 2024-12-04 released (on PhysicsX platform)A PhysicsX model that quickly generates aircraft shapes and predicts how they will perform aerodynamically and structurally.
  4. Ai.rplane 2024-12-04 released (free public app)A free public web app, built on LGM-Aero, that creates new aircraft designs and estimates their performance.
Leonard Circus, London (EC2A 4DQ)Companies House · physicsx.ai/newsroom/physicsx-announces-300m-ser · tech.eu/2026/06/08/physicsx-raises-300m-at-24bn- · uktech.news/funding/uk-tech-funding-roundup-this · physicsx.ai/newsroom/physicsx-raises-135m-seriesverified
Pleiades (Prima Mente epigenome foundation models)Chips, models and productsPrima Mente's family of epigenome foundation models (90M, 600M, 7B parameters) trained on DNA methylation data to detect Alzheimer's signals in blood. releasedKey facts 90M, 600M and 7B parameters; 1.9 trillion tokens (BioSpace release, 16 Jul 2025); 7B trained on 256 NVIDIA H200 GPUs; 0.89 accuracy for early Alzheimer's and Parkinson's detection, 0.97 with protein tests (company figures; NVIDIA case study, 2 Feb 2026); 32-GPU reserved Nebius H200 cluster raised pre-training throughput by 12%; plan to scale beyond 100B parameters (Nebius customer story, 2026)Made by Prima Mente · 2025-07-16Versions Pleiades → Pleiades-NeuronLondonnvidia.com/en-us/case-studies/primamente/ · biospace.com/press-releases/prima-mente-launcheslab's own pages
Pleiades-NeuronChips, models and productsPrima Mente model that finds very rare DNA fragments shed by brain cells in blood. research (technical report)Key facts Published 24 Oct 2026; built on Pleiades; reported higher sensitivity than a CpA heuristic baseline; found raised neuron-derived cell-free DNA after brain death, in Neuro-ICU patients and in measles patients; access: technical report (PDF)Made by Prima Mente · 2026-10-24Versions Pleiades → Pleiades-NeuronLondonprimamente.com/blog/pleiades-neuron/ · primamente.com/research/lab's own pages
PolyAICompaniesBuilds voice agents, running on its own dialogue models, that answer customer phone calls for hotels, banks, shops and other companies.
Profile
Legal name
PolyAI Ltd. (as named by SiliconANGLE; not seen on a company page)
Founders
Nikola Mrkšić — Co-founder & CEO; Shawn Wen — Co-Founder and CTO (other co-founders not named on pages opened)
Chief executive
Nikola Mrkšić — from null
Chief scientist or research lead
Shawn Wen — Co-Founder and CTO
Other leaders
Helen Gruel — Senior Vice President, Engineering; Morgan Norman — Chief Marketing Officer; Paul Asoyan — Vice President, Strategic Alliances & Business Development; Danielle Waknine — Vice President, Vertical Expansions (all announced 12 May 2026)
Hq
null on pages opened; other_offices: Toronto tech hub (first Canadian presence, May 2026)
Headcount
null (Toronto hub planned at about twenty people by end of Q2 2026)
What it does
Builds voice AI agents that answer customer phone calls for large companies — energy utilities, banks, casinos and hotels, restaurants, healthcare transport — using its own dialogue models and an agent-building platform.
Research focus
spoken-dialogue models that take audio in directly; voice agents for customer service
Models and products
Dialog-RSN-1 — voice dialogue model that takes audio directly; company-reported latency 280–500 ms — 30 Jul 2026; Wren — coding agent behind its dialogue agents (AI Magazine headline, 11 Sep 2026); Agentic Dialog Platform opened to self-serve (CMSWire headline, 18 May 2026); Dialog-Raven-3.5 — date not shown; Dialog Studio — date not shown
Customers or partners named
PG&E — calls during wildfires — Dec 2025 — https://poly.ai/blog/polyai-raises-86-million-series-d; UniCredit (Zagrebačka Banka) — NPS up 14 points — Dec 2025 — same url; Golden Nugget — Dec 2025 — same url; Caesars — Dec 2025 — https://siliconangle.com/2025/12/15/call-center-chatbot-startup-polyai-raises-86m-750m-valuation/; SafeRide Health — medical transport calls — Dec 2025 — https://poly.ai/blog/polyai-raises-86-million-series-d; Fogo de Chão — voice agent across all US restaurants — 5 Feb 2026 (Yahoo Finance headline on PolyAI media page) — https://poly.ai/in-the-media; Epic EHR — integration — 18 Aug 2026 (Destination CRM headline) — https://poly.ai/in-the-media
Papers and awards
Best AI Tool, 2026 Digiday Technology Awards — 22 Sep 2026 (headline listed on PolyAI media page)
Compute
Dialog-RSN targets sub-300 ms on A100 GPUs using an 8B dense or 30B sparse model (company-reported, via SiliconANGLE) — 30 Jul 2026
Funding total as published
over 200,000,000 USD — Dec 2025 — https://poly.ai/blog/polyai-raises-86-million-series-d
Valuation reported
750,000,000 USD — Dec 2025 — (trade) https://siliconangle.com/2025/12/15/call-center-chatbot-startup-polyai-raises-86m-750m-valuation/; same headline listed on https://poly.ai/in-the-media (single_source: no company statement of the figure opened)
Rounds found
2025-12 (blog dated 17 Dec 2025; trade 15 Dec 2025) — Series D — 86,000,000 USD — Georgian, Hedosophia, Khosla Ventures (co-leads) — NVentures, British Business Bank, Sands Capital, Squarepoint Ventures, Citi Ventures, Point72 Ventures
Uk government links
British Business Bank invested in the Series D (Dec 2025)
Status now
Launched its Dialog-RSN-1 voice model in July 2026 after an $86m Series D in December 2025; opened a Toronto hub in 2026.

Labs Round 7 deep profile (30 Oct 2026), as the researchers wrote it; confidence: verified (Series D amount: company + trade); valuation single_source (trade).

Sources

  1. poly.ai/blog/polyai-raises-86-million-series-d
  2. poly.ai/blog/polyai-makes-four-senior-hires
  3. poly.ai/in-the-media
Models and products (6)
  1. PolyAI Agentic Dialog Platform (open to builders: Poly Agent Builder and ADK) 2026-05-18 released (self-serve)PolyAI's platform for building enterprise voice agents, opened to all builders with an agent builder and developer kit.
  2. Dialog-RSN-1 — (exact release date not stated; API live on post of 18 May 2026, page modified 25 Sep 2026) released (API)PolyAI's newest dialogue model, offered by API, which the company says handles conversational turn-taking differently.
  3. Concierge Chat (Omnichannel Intelligence Hub) 2026-01-07 live (product)PolyAI's chat assistant for retail websites, replacing search boxes with a single guided conversation.
  4. Owl (PolyAI speech recognition model) 2025-06-03 live (inside PolyAI platform)PolyAI's own speech-to-text model, built to hear callers accurately over phone lines with noise and accents.
  5. Raven (PolyAI's conversational model family; current Raven 3.5) no date released (default model on PolyAI platform)PolyAI's own large language model for live customer-service conversations by phone and chat.
  6. Wren Dialog Agent — (date not stated) live (product)PolyAI agent for designing, building, testing and improving voice and chat agents.
Sheldon Square, London (W2 6HY)Companies House · poly.ai/blog/polyai-raises-86m-to-transform-how- · poly.ai/blog/polyai-raises-50-million-series-c · find-and-update.company-information.service.gov. · find-and-update.company-information.service.gov.verified
Prima DaedalusChips, models and productsPrima Mente's multi-agent AI workflow that reprocesses published biology experiments into checked training data and tests for its models. research (code not yet public)Key facts Published 24 Sep 2026; three-stage workflow with human approval gates; in one test recovered 41 of 79 author-reported exon-cell-type pairs; code stated to be released "very soon"; access: paper onlyMade by Prima Mente · 2026-09-24Londonprimamente.com/blog/prima-daedalus/ · primamente.com/research/lab's own pages
Prima MenteCompaniesAI biology company that generates its own data and builds biological foundation models, starting with Alzheimer's, Parkinson's and other brain diseases.
Profile
Founded
2023 (Tech Funding News, trade, single)
Founders
Hannah Madan — co-founder; Ravi Solanki — Co-Founder; Jonathan Wan — co-founder (names from Tech Funding News; Solanki's title from GOV.UK)
Chief executive
null (no CEO title found)
Hq
London (England: London); other_offices: San Francisco; United Arab Emirates
Headcount
null (team section of the website names 35 people, October 2026; not a stated headcount)
What it does
An AI biology company that builds foundation models from the human epigenome and brain-related data to find early signs, subtypes and drug targets of brain diseases, starting with Alzheimer's, Parkinson's and ALS. It generates some of its own data and works with NHS and university studies.
Research focus
biological foundation models; epigenome; neurodegenerative disease (Alzheimer's, Parkinson's, ALS); biomarkers; interpretability (with Goodfire)
Models and products
Pleiades — whole-epigenome foundation models (90M, 600M and 7B parameters; 1.9 trillion tokens) — 16 July 2025; Pleiades 2 — model combining five biological data modalities, in development on Isambard-AI — June 2026; Prima Daedalus — multi-agent system for training/evaluation data — 24 September 2026; Pleiades-Neuron — detection of neuronal cell-free DNA in blood — 24 October 2026
Customers or partners named
Goodfire — interpretability study of Alzheimer's biomarkers — 28 January 2026 — https://www.primamente.com/research/; NVIDIA and Nebius — case study (0.89 accuracy, early-stage Alzheimer's and Parkinson's from blood) — 2 February 2026 — https://www.primamente.com/research/; University of Oxford (READOUT study, NIHR-supported; Prima Mente industry partner) — date null — https://www.primamente.com/research/; Oxford, Imperial and Edinburgh — research collaborations — 16 April 2026 — https://www.gov.uk/government/news/ai-firms-pioneering-drug-discovery-cheaper-supercomputing-and-more-get-first-backing-through-uks-sovereign-ai; collaborators listed on site: Johnson & Johnson, Salk Institute, Dementias Platform UK, Office for Life Sciences, Stanford University and others — October 2026 — https://www.primamente.com/
Papers and awards
Longitude Prize on ALS Discovery Award winner — 7 May 2026; "Detecting ultra-rare neuronal cell-free DNA in blood with Pleiades-Neuron" — 24 October 2026; "Automating training and ground truth data for BioAI models and evals" — 24 September 2026
Compute
AIRR allocation through Sovereign AI (April 2026); developing Pleiades 2 on Isambard-AI; "nearly 3x speedups" with NVIDIA Blackwell GPUs (telecomtv, June 2026); Nebius reports 1.19 million tokens per second on 16 nodes (via primamente.com research page)
Funding total as published
1,400,000 USD ("Total VC raised") — as of 16 April 2026 — https://techfundingnews.com/uks-sovereign-ai-fund-names-7-startups-its-supporting-in-its-first-batch-here-are-the-details/ (trade, single); NVIDIA planned investment October 2026, amount not disclosed
Rounds found
investors named by Tech Funding News: Beyond Capital, Bluebirds Capital, Episode 1 Ventures, Antler, Octopus Ventures (no dates or amounts); 2026-10-13 — strategic investment (planned) — undisclosed — NVIDIA (already DEAL-LAB-029 / DEAL-LAB-2003)
Uk government links
Sovereign AI / AIRR compute, one of six firms — 16 April 2026 (already DEAL-LAB-2035); Office for Life Sciences listed as collaborator; SANDBOX study across 21 NHS sites (company research page)
Status now
Active London company with offices in San Francisco and the UAE; NVIDIA announced a planned minority investment on 13 October 2026.

Labs Round 7 deep profile (30 Oct 2026), as the researchers wrote it; confidence: verified (location, models, government compute, NVIDIA); founders list, founding year and funding total single_source.

Sources

  1. primamente.com/
  2. primamente.com/blog/pleiades-july-2025/
  3. primamente.com/research/
  4. primamente.com/news/
  5. gov.uk/government/news/ai-firms-pioneering-drug-
  6. sovereignai.gov.uk/post/our-first-investments
  7. telecomtv.com/content/digital-platforms-services
  8. techfundingnews.com/uks-sovereign-ai-fund-names-
  9. blogs.nvidia.com/blog/uk-ecosystem-deployment/
Models and products (3)

Version chainsPleiades → Pleiades-Neuron

  1. Pleiades-Neuron 2026-10-24 research (technical report)Prima Mente model that finds very rare DNA fragments shed by brain cells in blood.
  2. Prima Daedalus 2026-09-24 research (code not yet public)Prima Mente's multi-agent AI workflow that reprocesses published biology experiments into checked training data and tests for its models.
  3. Pleiades (Prima Mente epigenome foundation models) 2025-07-16 releasedPrima Mente's family of epigenome foundation models (90M, 600M, 7B parameters) trained on DNA methylation data to detect Alzheimer's signals in blood.
Londonprimamente.com/ · blogs.nvidia.com/blog/uk-ecosystem-deployment/ · primamente.com/news/ · find-and-update.company-information.service.gov.verified
Project AstraChips, models and productsGoogle DeepMind's prototype universal assistant that sees and hears through a phone camera or prototype glasses and talks back in real time. research prototype (trusted testers)Key facts Used by a limited number of trusted testers (waitlist); version for blind and low-vision users in a trusted-tester programme; runs on Android phones and prototype glasses; features such as screen sharing and video understanding moved into Gemini LiveMade by Google DeepMind · — (date not stated on the project page)Londondeepmind.google/models/project-astra/one source
Real World Model (Stanhope AI)Chips, models and productsStanhope AI's on-device AI framework based on active inference, a neuroscience theory, being tested in autonomous drones and robots. in testing (drones and robotics)Key facts Draws on the Free Energy Principle and active inference; lean models that run on-device; tested with unnamed international partners in drones and robotics (Paladin Capital, 12 Feb 2026); $8m seed led by Frontline Ventures, 12 Feb 2026 (existing DEAL-LC-074)Made by Stanhope AI · 2026-02-12Londonpaladincapgroup.com/stanhope-ai-raises-8m-as-rea · siliconrepublic.com/start-ups/stanhope-ai-co-foulab's own pages
Recraft V2 (Recraft 20B)Chips, models and productsRecraft's image model pitched as the first built for designers, with about 20 billion parameters. superseded (product)Key facts Announced 13 Mar 2024; about 20 billion parameters (post compares SDXL and Playground v2.5 at 2.6 billion each); user study on 1,632 PartiPrompts with 1,435 assessors, stated comparable to DALL-E 3 and ahead of Midjourney v6; access: productMade by Recraft · 2024-03-13Versions Recraft V2 → V3 → V4 → V4.1 → V4.1 FlashLondonrecraft.ai/blog/recraft-v2one source
Recraft V4.1 FlashChips, models and productsRecraft's fastest image model, producing a picture about 1.3 seconds after the prompt. released (product and API)Key facts Announced 23 Sep 2026; about 1.3 seconds median from prompt to image; in Recraft's 100-prompt test (21 Sep 2026) median 1.4 s and p95 1.8 s; "Enhance" regenerates with V4.1 Pro at 2048x2048; access: product and APIMade by Recraft · 2026-09-23Londonrecraft.ai/blog/meet-recraft-v4-1-flashone source
RecurrentGemma (2B and 9B)Chips, models and productsOpen Gemma language model built on a different design (Griffin) that mixes recurrence with local attention to run more efficiently. released (open weights)Key facts Initial release 9 Apr 2024; 9B variant released 11 Jun 2024; based on the Griffin architecture (arXiv 2402.19427); access: open weightsMade by Google DeepMind · 2024-04-09Versions Gemma → Gemma 2 (and CodeGemma, RecurrentGemma, PaliGemma → PaliGemma 2, ShieldGemma → ShieldGemma 2, DataGemma)Londonai.google.dev/gemma/docs/recurrentgemma · ai.google.dev/gemma/docs/releaseslab's own pages
Recursive FlintChips, models and productsRecursive Superintelligence's first public model: a 40-billion-parameter coding model built to run fast on local machines. releasedKey facts 40B total parameters, about 2B active, mixture-of-experts (recursive.com, read 2026-10-30); the company says its automated research system shaped every stage of the build; listed as "Model 01"Made by Recursive Superintelligence · 2026 (date not given on page)Londonrecursive.com/ · recursive.com/one source
Recursive SuperintelligenceCompaniesResearch company building AI systems that improve themselves, starting by automating AI research and the scientific method.
Profile
Founded
null (emerged from stealth 13 May 2026)
Founders
Richard Socher — leads the company "alongside seven co-founders" (former chief scientist at Salesforce, founder of You.com); Yuandong Tian — co-founder (former research scientist director, Meta FAIR); Tim Rocktäschel — co-founder (Professor of AI, UCL; former principal scientist, Google DeepMind); Alexey Dosovitskiy — co-founder; Josh Tobin — co-founder (formerly OpenAI); Caiming Xiong — co-founder; Tim Shi — co-founder; Jeff Clune — co-founder
Chief executive
Richard Socher — leads the company (title not stated in source)
Other leaders
Peter Norvig — adviser
Hq
London (England: London) per BusinessCloud; other_offices: San Francisco
Headcount
more than 25 — May 2026 — BusinessCloud and TNW (both trade)
What it does
AI research company building systems that run their own experiments to improve themselves safely — "recursive self-improvement" through open-ended automated scientific discovery.
Research focus
recursive self-improvement; automated scientific discovery
Compute
null (NVIDIA and AMD are investors)
Funding total as published
650,000,000 USD (£480m) — May 2026 — trade
Valuation reported
4,650,000,000 USD — 14 May 2026 — (trade) TNW; about £3.5bn — 14 May 2026 — (trade) BusinessCloud (two trade sources; no primary opened)
Rounds found
2026-05-13 — first round (TNW: emerged from stealth; other listings call it Series A — not verified) — 650,000,000 USD / £480m — GV, Greycroft — NVIDIA, AMD (Ventures)
Status now
Emerged from stealth in May 2026 with $650m at a reported $4.65bn valuation; offices in London and San Francisco.

Labs Round 7 deep profile (30 Oct 2026), as the researchers wrote it; confidence: single_source (two trade sources, no primary); sources are listed in the record panel.

Models and products (1)
  1. Recursive Flint 2026 (date not given on page) releasedRecursive Superintelligence's first public model: a 40-billion-parameter coding model built to run fast on local machines.
King's Cross, London (N1 9FN)Companies House · tech.eu/2026/05/13/recursive-superintelligence-e · find-and-update.company-information.service.gov. · find-and-update.company-information.service.gov. · recursive.com/one source
RETRO (Retrieval-Enhanced Transformer)Chips, models and productsLanguage model that looks up passages in a very large text database while it writes, so a smaller model can match much larger ones. research (paper only)Key facts Announced 8 Dec 2021; 7.5B-parameter model with a retrieval database of up to 2 trillion tokens; outperformed 175B Jurassic-1 on 10 of 16 Pile datasets and 280B Gopher on 9 of 16; access: paper onlyMade by Google DeepMind · 2021-12-08Londondeepmind.google/blog/improving-language-models-bone source
RoboCatChips, models and productsSelf-improving robot agent that learns new tasks on different real robot arms from a few hundred demonstrations. research (paper only)Key facts Announced 20 Jun 2023; learns a new task from 100 to 1,000 demonstrations; data from four robot types; success on unseen tasks after 500 demonstrations rose from 36% (initial) to 74% (final); arXiv 2306.11706; access: paper onlyMade by Google DeepMind · 2023-06-20Londondeepmind.google/blog/robocat-a-self-improving-roone source
SDXL (Stable Diffusion XL)Chips, models and productsLarger Stable Diffusion model for higher-resolution images, with a second text encoder and an optional refining model. superseded (open weights)Key facts Paper arXiv 2307.01952 submitted 4 Jul 2023; UNet backbone three times larger than earlier versions; second text encoder; separate refinement model; SDXL 1.0 announcement page not opened; access: open weightsMade by Stability AI · 2023-07-04 (paper)Versions Stable Diffusion v1 → 2.0 → SDXL → SDXL Turbo → SD3 → SD3.5Londonarxiv.org/abs/2307.01952one source
SDXL Turbo (Adversarial Diffusion Distillation)Chips, models and productsFast version of SDXL that makes an image in a single step, close to real time. released (non-commercial open weights)Key facts Steps cut from 50 to one; 512x512 image in 207 ms on an A100 (67 ms for one UNet pass); method paper arXiv 2311.17042 submitted 28 Nov 2023; access: weights and code on Hugging Face under non-commercial research licence, demo on Clipdrop; listed among Stability Core Models (page updated 20 May 2026)Made by Stability AI · 2023-11-28Versions Stable Diffusion v1 → 2.0 → SDXL → SDXL Turbo → SD3 → SD3.5Londonstability.ai/news-updates/stability-ai-sdxl-turb · arxiv.org/abs/2311.17042lab's own pages
Sessions (Roleplay, Survey and Expert Sessions)Chips, models and productsSynthesia platform where interactive AI avatars hold live conversations with staff for coaching, surveys and guidance. live (Roleplay and Survey); Expert comingKey facts Roleplay Sessions introduced 22 Jul 2026; Sessions platform launched 1 Oct 2026; 78% of learners returned to retry a roleplay and 83% of those improved their score (company figures); Expert Sessions coming soon; Hays partnership on Roleplay Sessions 22 Oct 2026Made by Synthesia · 2026-10-01Londonsynthesia.io/post/sessions-interactive-avatar-pl · synthesia.io/blog/category/synthesialab's own pages
ShieldGemmaChips, models and productsOpen Gemma 2 models that check text going into and out of AI systems for harmful content. released (open weights)Key facts Initial release 31 Jul 2024; sizes 2B, 9B and 27B; built on Gemma 2; targets four harm types (sexually explicit, dangerous content, hate, harassment); paper arXiv 2407.21772; access: open weightsMade by Google DeepMind · 2024-07-31Versions Gemma → Gemma 2 (and CodeGemma, RecurrentGemma, PaliGemma → PaliGemma 2, ShieldGemma → ShieldGemma 2, DataGemma)Londonai.google.dev/gemma/docs/shieldgemma/model_card · ai.google.dev/gemma/docs/releaseslab's own pages
ShieldGemma 2Chips, models and productsOpen model that checks images for harmful content against set safety policies. released (open weights)Key facts Released 10 Mar 2025; 4B-parameter image safety model built on Gemma 3; access: open weightsMade by Google DeepMind · 2025-03-10Versions Gemma → Gemma 2 (and CodeGemma, RecurrentGemma, PaliGemma → PaliGemma 2, ShieldGemma → ShieldGemma 2, DataGemma)Londonai.google.dev/gemma/docs/releases · deepmind.google/models/gemma/shieldgemma-2/lab's own pages
Snowdon (Frontier AI Research Lab model)Chips, models and productsA re-aligned version of open Qwen models built by the Thomson Reuters–Imperial lab and used as the base for Thomson. released (used as Thomson base model)Key facts Snowdon-1.0-Large and Snowdon-1.1-Small, value-realigned versions of Qwen3.5-397B and Qwen3.6-35B (technical report, 2026)Made by Frontier AI Research Lab (Thomson Reuters and Imperial College London) · 2026-08Londonthomsonreuters.com/content/dam/ewp-m/documents/t · thomsonreuters.com/en-us/posts/innovation/how-welab's own pages
Sparrow (dialogue agent)Chips, models and productsResearch chatbot trained with human feedback and rules to give evidence-backed answers and avoid unsafe replies. research (paper only)Key facts Announced 22 Sep 2022; gave a plausible answer supported by evidence 78% of the time on factual questions; participants tricked it into breaking rules 8% of the time; paper arXiv 2209.14375; access: paper onlyMade by Google DeepMind · 2022-09-22Londondeepmind.google/blog/building-safer-dialogue-ageone source
Speech to Speech voice changer models (English STS v1 and v2; Multilingual STS v2)Chips, models and productsElevenLabs models that turn one person's recorded speech into another voice while keeping the delivery. released (API)Key facts Model ids eleven_english_sts_v1, eleven_english_sts_v2 (English only) and eleven_multilingual_sts_v2 (29 languages); 10,000-character limit listed for English STS; ElevenLabs blogged on voice conversion research on 6 Sep 2022; access: APIMade by ElevenLabs · — (release dates not stated)Londonelevenlabs.io/docs/overview/models · elevenlabs.io/blog/voice-conversionone source
Stable Diffusion (first public release, v1)Chips, models and productsThe original open text-to-image model that made pictures from typed prompts on an ordinary consumer graphics card. superseded (open weights)Key facts Runs on under 10 GB of VRAM; 512x512 images in a few seconds; trained on LAION-Aesthetics, a subset of LAION 5B; over 10,000 beta testers creating 1.7 million images a day; the post credits CompVis latent diffusion work, LAION and EleutherAI, and Runway's Patrick Esser as project lead; access: open weights (code on GitHub, model card on Hugging Face)Made by Stability AILondonstability.ai/news-updates/stable-diffusion-annouone source
Stable Diffusion 2.0Chips, models and productsSecond major Stable Diffusion release, with a new text encoder, higher default resolution and a 4x upscaler. superseded (open weights)Key facts Default resolutions 512x512 and 768x768; upscaler raises resolution 4x; OpenCLIP text encoder (LAION, supported by Stability AI); trained on an aesthetic subset of LAION-5B filtered for adult content; access: open source plus Stability API and DreamStudioMade by Stability AILondonstability.ai/news-updates/stable-diffusion-v2-reone source
Stable Zero123 (and Stable Zero123C)Chips, models and productsModel that generates new views of an object from a single picture, as a step to making 3D objects. released (non-commercial; Zero123C commercial for members)Key facts Training about 40 times faster than Zero123-XL; based on Stable Diffusion 1.5; trained on a filtered subset of Objaverse; 24 GB VRAM recommended for 3D generation; Stable Zero123C (CC-BY and CC0 objects only) usable commercially with a Stability membershipMade by Stability AI · 13 Dec 2023 (post updated 8 Jan 2024)Londonstability.ai/news-updates/stable-zero123-3d-geneone source
StableLM (Alpha, 3B and 7B)Chips, models and productsStability AI's first suite of open language models, released in two small sizes. superseded (open weights)Key facts Alpha models of 3 billion and 7 billion parameters, with 15 to 65 billion stated to follow; trained on 1.5 trillion tokens of content; base models CC BY-SA-4.0, fine-tuned research models CC BY-NC-SA 4.0; access: open weights (GitHub); successor Stable LM 2 already listed (BIZ-LAB-7039)Made by Stability AIVersions StableLM → Stable LM 2Londonstability.ai/news-updates/stability-ai-launches-one source
SynthesiaCompaniesMakes software that turns typed text into videos presented by digital human avatars, mostly for company training.
Profile
Founded
2017
Founders
Victor Riparbelli — CEO & Co-Founder; Steffen Tjerrild — COO & Co-Founder; Lourdes Agapito — co-founder (role not stated on page)
Chief executive
Victor Riparbelli — from null
Other leaders
Peter Hill — CTO; Ken Lawshe — CRO; Laura Gonzalez — Chief of Staff and Head of People; Gabe Stern — General Counsel; Joe Ciarallo — Head of Communications, United States; Brett Taylor — hired with Seattle expansion, 22 Jun 2026 (role not opened)
Hq
London (England: London) — 20 Triton Street, Regent's Place, NW1 3BF; other_offices: New York; Austin and Berlin (new, April 2026); Paris and Zurich (expanded, April 2026); Seattle (expansion, June 2026)
Headcount
careers page shows both "600+ GLOBAL EMPLOYEES" and "900+ GLOBAL EMPLOYEES" (undated, read 30 Oct 2026) — https://www.synthesia.io/careers; plan to grow global headcount by more than 70% in 2026 (21 Apr 2026). UK headcount not published; "roughly 80%" of London HQ staff are immigrants (careers page).
What it does
AI video platform for companies. Users type text and get training, sales or internal videos presented by AI avatars and voices in many languages. Newer products add interactive avatars that coach, interview and guide people, and a tool that makes video and motion graphics from a single prompt.
Research focus
AI avatars, video generation, dubbing, interactive video agents
Models and products
Syren — video and motion graphics from a single prompt — 9 Oct 2026; Sessions — platform where interactive avatars coach, interview and guide — 1 Oct 2026; AI Press Officer — 26 Aug 2026; Roleplay Sessions — 22 Jul 2026; Interactive Avatar API — 15 Jul 2026; Express-3 avatar model and AI Assistant — 15 Jul 2026; Dubbing 2.0 — 15 Jul 2026; Next-Gen Pronunciation Controls — 11 Jun 2026
Customers or partners named
Hays — Roleplay Sessions for skills development — 22 Oct 2026 — https://www.synthesia.io/blog/category/synthesia; NVIDIA — interactive avatar of its CEO at GTC Berlin — 20 Oct 2026 — same url; Nexity — customer quoted on production timelines — 21 Apr 2026 — https://www.synthesia.io/post/synthesia-global-expansion-austin-berlin-paris-zurich-2026; KnowBe4 — partnership — 20 Apr 2026 — https://www.synthesia.io/blog/category/synthesia; Multiverse — 3 Jun 2026 — same url; Cinder — trust and safety partner — 4 Jun 2026 — same url; AWS — 17 Sep 2025 — same url; "over 90% of the Fortune 100" — 26 Jan 2026 — https://www.synthesia.io/post/series-e-200-million-4-billion-valuation-future-work
Papers and awards
Fast Company "Next Big Things in Tech" list — 6 Oct 2026; signed the EU Article 50 Code of Practice — 29 Jul 2026
Valuation reported
4,000,000,000 USD — 26 Jan 2026 — https://www.synthesia.io/post/series-e-200-million-4-billion-valuation-future-work (single source)
Rounds found
2026-01-26 — Series E — 200,000,000 USD — GV (lead) — Evantic, Hedosophia, NVentures, Accel, Kleiner Perkins, NEA, PSP Growth, Air Street Capital, FirstMark, MMC Ventures; plus employee secondary sale run with Nasdaq at the same valuation; 2025-04-15 — strategic investment — undisclosed — Adobe Ventures (ARR passed $100m)
Uk government links
Memorandum of Understanding with the UK Government "to advance AI adoption and upskilling across the UK" — 9 Jun 2026
Status now
Valued at $4bn after its January 2026 Series E; expanding offices in the US and Europe and launching interactive-avatar products.

Labs Round 7 deep profile (30 Oct 2026), as the researchers wrote it; confidence: single_source.

Sources

  1. synthesia.io/post/series-e-200-million-4-billion
  2. synthesia.io/post/series-e-200-million-4-billion
  3. synthesia.io/careers
  4. synthesia.io/post/synthesia-global-expansion-aus
  5. synthesia.io/blog/category/synthesia
  6. synthesia.io/post/100-million-revenue-adobe-inve
Models and products (10)

Version chainsEXPRESS-1 (Apr 2024) → Express-2 (named with Synthesia 3.0, 1 Oct 2025; action avatars 13 Nov 2025) → Express-3 (15 Jul 2026)

  1. Personal Avatars (Synthesia digital twins) — (launch date not stated; docs page modified 22 Oct 2026) live (paid plans)Synthesia feature that builds a talking digital copy of a real person from a photo, webcam or studio footage.
  2. Syren (Synthesia prompt-to-video tool) 2026-10-09 betaSynthesia's tool that turns a written request or documents into a finished business video with motion graphics, avatars, voice and music.
  3. Sessions (Roleplay, Survey and Expert Sessions) 2026-10-01 live (Roleplay and Survey); Expert comingSynthesia platform where interactive AI avatars hold live conversations with staff for coaching, surveys and guidance.
  4. Express-3 2026-07-15 released (rolling out to all plans)Synthesia's newest avatar model, with more natural lip sync, body movement and gestures that match the script.
  5. Interactive Avatar API 2026-07-15 released (enterprise; managed version planned)An API that lets developers put a live talking Synthesia avatar into their own website or app.
  6. Dubbing 2.0 (Synthesia) 2026-07-15 live (product)Synthesia's upgraded video dubbing, translating speech into other languages with matching lip movements.
  7. Express-2 (customisable action avatars) 2025-11-13 released (product)Synthesia's second avatar model, which lets one avatar both explain and act out a short clip in the same editor.
  8. Synthesia 3.0 and Video Agents 2025-10-01 released (platform)Synthesia's 2025 platform release, adding Video Agents that hold real-time conversations inside videos.
  9. Synthesia 2.0 2024-06-23 released (platform)Synthesia's rebuilt business platform for making and sharing AI videos, with personal avatars and an AI video assistant.
  10. EXPRESS-1 (Expressive Avatars) 2024-04-22 released (product)Synthesia's avatar model that delivers a script with matching tone, body language and lip sync.
20 Triton Street, Regent's Place, London (NW1 3BF)Companies House · synthesia.io/post/series-e-200-million-4-billion · uktech.news/ai/synthesia-4bn-valuation-confirmed · synthesia.io/post/synthesia-secures-180m-in-seri · synthesia.io/newsroomverified
Text to Sound Effects v2 (eleven_text_to_sound_v2)Chips, models and productsElevenLabs model that makes sound effects from a written description. released (API; default model)Key facts Remains the default sound-effects model; SFX v3 (BIZ-LAB-7036) has limited access; access: APIMade by ElevenLabs · — (release date not stated)Londonelevenlabs.io/docs/overview/modelsone source
TR Labs Foundational Research, London hub (Thomson Reuters)Data centres and compute sitesThomson Reuters' model-training research team, built from the London start-up Safe Sign Technologies, with London one of its research hubs.Key facts Team "grew" from a 15-person start-up to a 50-person global group (TR, 12 Dec 2025). Research hubs in London, Toronto, Zug and Minneapolis (TR, 15 Dec 2025). Planned to bring the first Thomson Reuters large language model to market in 2026 (TR, 12 Dec 2025). Model "Thomson" (table: "Thomson-1-Large"): early benchmark results published 31 Jul 2026 — PrBench Legal Hard 0.352, Instruction Following 0.914, Long Context 0.753; parameter count not disclosed; to become the default model for Tabular Analysis in CoCounsel Legal, first integration August 2026 (https://www.thomsonreuters.com/en-us/posts/innovation/thomson-reuters-built-its-own-ai-model-that-now-ranks-among-the-worlds-best/). Head: Dr Jonathan Schwarz, Head of Foundational Research.
Profile
Owner or parent
Thomson Reuters (TR Labs)
Founded or opened
built from Safe Sign Technologies after Thomson Reuters bought it in Aug 2024; "now called Foundational Research" (Dec 2025)
Lab director or lead
Jonathan Schwarz — Head of Foundational Research (Dec 2025) and Head of AI Research, Thomson Reuters (Dec 2025; Jul 2026); co-founder of Safe Sign; previously seven years as a senior researcher at Google DeepMind
Other leaders
Joel Hron — Chief Technology Officer, Thomson Reuters; Wassim Seifeddine — TR Labs researcher
Site and address as published
London is listed among TR research hubs (London, Toronto, Zug, Minneapolis); the team's own base is not stated on pages opened
Headcount
"50-person global group" (grown from a "15-person startup") — 12 Dec 2025 — https://www.thomsonreuters.com/en-us/posts/innovation/from-startup-vision-to-ai-breakthrough-one-year-with-safe-sign/ (single source)
What it does
Thomson Reuters' in-house AI research group, built from London legal-AI start-up Safe Sign Technologies, which it bought in 2024. It trains and evaluates language models for legal and professional work and publishes on data mixtures, model efficiency and evaluation. It built "Thomson", Thomson Reuters' own model, which became the default model for Tabular Analysis in CoCounsel Legal from August 2026.
Research groups or focus
"data mixture optimization, model efficiency, and responsible evaluation methods"
Notable outputs
Thomson (Thomson-1-Large in the benchmark table) — Thomson Reuters' own model for professional and legal work, built on an open-source foundation model and trained on Westlaw, Practical Law, Checkpoint and Reuters content; default model for Tabular Analysis in CoCounsel Legal from Aug 2026 — announced 31 Jul 2026. The company reports it competitive with Claude Opus 4.8 and ahead of GPT-5.5 and Gemini 3.1 Pro on its own early tests.
Selected papers
two NeurIPS 2025 publications (titles not given on the page): data weighting that improves training efficiency "by as much as 500 percent", and evaluation methods that cut benchmarking cost "up to 18 times" — NeurIPS — Dec 2025 — https://www.thomsonreuters.com/en-us/posts/innovation/from-neurips-2025-how-thomson-reuters-is-advancing-the-frontier-of-ai-research/
Partners named
Imperial College London — Frontier AI Research Lab — 2 Dec 2025
Status now
Active: Thomson Reuters launched its own model, Thomson, in summer 2026.

Labs Round 8 deep profile (31 Oct 2026), as the researchers wrote it; confidence: verified (Schwarz's role appears on three Thomson Reuters pages and at UKTN); headcount and model claims one source.

Sources

  1. thomsonreuters.com/en-us/posts/innovation/from-n
  2. thomsonreuters.com/en-us/posts/innovation/from-s
  3. thomsonreuters.com/en-us/posts/innovation/thomso
Models and products (1)
  1. Thomson (Thomson Reuters large language model) 2026-08 in productionThomson Reuters' first in-house large language model, tested on legal, tax and news tasks and able to run inside a customer's own systems.
London (one of four hubs)thomsonreuters.com/en-us/posts/innovation/from-n · thomsonreuters.com/en-us/posts/innovation/from-s · thomsonreuters.com/en-us/posts/our-purpose/thoms · thomsonreuters.com/en-us/posts/innovation/how-weone source
Unlikely AICompaniesBuilds its own AI for decisions in regulated fields such as financial services, accounting and insurance, where answers must be explainable and correct.
Profile
Legal name
null (footer "© 2026 UnlikelyAI")
Founders
William Tunstall-Pedoe — founder and CEO
Chief executive
William Tunstall-Pedoe — from null
Other leaders
Tom Mason — senior hire, former CTO of Stability AI (title at UnlikelyAI not stated) — announced 1 September 2025; Amelia Armour — board member (Partner, Amadeus Capital Partners) — 2022; Berkowitz (first name not given) — board member — 2022; Rishy Shukla — Applied Science Manager; Ailish McLaughlin — Solutions Lead
Hq
null (described as a "U.K. foundation model maker"); other_offices: null
What it does
A UK company building "neurosymbolic" AI for decisions where errors are costly. It turns documents and rules written in plain language into a structured form, so answers can be explained and repeated exactly. It sells to financial services (customer chat guardrails), accounting (disclosure checklists) and insurance (routine claims).
Research focus
neurosymbolic AI; explainable and consistent decision-making; guardrails for LLMs
Models and products
UnlikelyAI platform — decisioning for fintech, accounting and insurance — 2026 (site); research site lab.unlikely.ai
Customers or partners named
Lloyds Banking Group — testimonial from Ranil Boteju, Chief Data and Analytics Officer — 2026 — https://unlikely.ai/
Papers and awards
National AI Awards 2026 — AI Innovation of the Year (UnlikelyAI) and Alan Turing Innovator of the Year (William Tunstall-Pedoe FREng) — July 2026
Funding total as published
20,000,000 USD seed (plus a £1.1m angel round in October 2020) — as of 15 September 2022 — https://unlikely.ai/unlikely-ai-raises-20m-in-oversubscribed-seed-round
Rounds found
2020-10 — angel — 1,100,000 GBP — none (26 angel investors) — (NEW: DEAL-LAB-9013); 2022-09-15 — seed — 20,000,000 USD — Amadeus Capital Partners and Octopus Ventures (co-leads) — Cambridge Innovation Capital, Metaplanet (Jaan Tallinn) and others (already DEAL-LAB-030)
Status now
Active; won two National AI Awards in July 2026 and posted company news up to 29 October 2026.

Labs Round 7 deep profile (30 Oct 2026), as the researchers wrote it; confidence: single_source (all from company pages; seed round already verified in the deal file).

Sources

  1. unlikely.ai/unlikely-ai-raises-20m-in-oversubscr
  2. unlikely.ai/newsroom/ai-awards-2026
  3. unlikely.ai/
  4. unlikely.ai/newsroom/alexa-co-creator-gives-firs
  5. unlikely.ai/newsroom
Models and products (1)
  1. UnlikelyAI neurosymbolic platform 2026-07-22 in useUnlikelyAI's platform joining neural models with symbolic logic so banks, insurers and accountants can check AI decisions.
LondonCompanies House · unlikely.ai/ · uktech.news/ai/unlikely-ai-funding-20220915 · find-and-update.company-information.service.gov. · find-and-update.company-information.service.gov.verified
V2A (video-to-audio research)Chips, models and productsResearch system that generates soundtracks, effects and dialogue to match the action in a video. research (not released)Key facts Announced 17 Jun 2024; not opened to the public pending safety assessments and testing; access: none (research)Made by Google DeepMind · 2024-06-17Londondeepmind.google/blog/generating-audio-for-video/one source
Veo (first version) in YouTube Dream ScreenChips, models and productsGoogle DeepMind's first Veo video model, brought to YouTube Shorts to make short background clips from a chosen image. superseded (product integration)Key facts 18 Sep 2024: Dream Screen uses Imagen 3 to make four candidate images, then Veo generates a 6-second background video; 6-second standalone clips stated for 2025; successors Veo 2 and Veo 3 already listed (BIZ-LAB-7163, 7166)Made by Google DeepMind · 2024-09-18 (YouTube integration announced; May 2024 launch post on blog.google not opened)Londondeepmind.google/blog/empowering-youtube-creatorsone source
Voice Design (first generative voice model, in Voice Lab)Chips, models and productsElevenLabs model that creates new synthetic voices from scratch rather than copying a real speaker. superseded (product)Key facts Announced 11 Jan 2023; "Design Voice" scheduled for February 2023 inside Voice Lab on the beta platform; successor Voice Design v3 already listed (BIZ-LAB-7027); access: productMade by ElevenLabs · 2023-01-11Londonelevenlabs.io/blog/enter-the-new-year-with-a-banone source
Waymo (London and Oxford engineering hubs)CompaniesUS self-driving company whose first engineering hubs outside the US are in London and Oxford; it plans driverless taxi rides in London. active; London driverless ride-hailing planned, commercial rides not yet available (Waymo UK FAQ)Key facts London and Oxford named as "first international engineering hubs", teams on large-scale closed-loop simulation (15 Oct 2025); intends driverless rides in London "in 2026" (15 Oct 2025); "further investing in our UK team" on the AI architecture of the Waymo Driver (14 Apr 2026); test vehicles with trained specialists at the wheel driven "across tens of thousands of miles on London roads" (14 Apr 2026); "tens of millions of pounds" invested locally in London vehicle operations and facilities (24 Sep 2026); service to be supported by "hundreds of jobs in the UK" (24 Sep 2026); fleet partner Moove; vehicle partner Jaguar Land Rover; permits under DfT Automated Passenger Services scheme with TfL consent
Profile
Owner or parent
Waymo
Founded or opened
hubs named on 15 Oct 2025 as Waymo's "first international engineering hubs"; opening date not stated
Lab director or lead
Not stated for the hubs. Tekedra Mawakana — co-CEO, Waymo
Site and address as published
London; Oxford (no addresses)
Headcount
"hundreds of jobs in the UK" to support the London service — 24 Sep 2026 — waymo.com (single source)
What it does
Waymo's first engineering hubs outside the US, in London and Oxford, whose teams work on large-scale closed-loop simulation for the Waymo Driver. Waymo plans fully driverless ride-hailing in London under the Automated Passenger Services permit scheme. That needs approval from the Department for Transport and consent from Transport for London.
Research groups or focus
"large-scale, closed-loop simulation" (Oct 2025)
Notable outputs
Not stated (none credited to the UK hubs)
Funding or budget as published
"tens of millions of pounds" invested locally in London vehicle operations and facilities — 24 Sep 2026 — https://waymo.com/blog/2026/09/visionforlondon/ (single source)
Partners named
Moove — fleet operations — Oct 2025 (already DEAL-LAB-5027); Jaguar Land Rover — all-electric I-PACE vehicles — Oct 2025; London Cycling Campaign, Royal National Institute of Blind People, Women in Transport — named in the Vision for London — Sep 2026
Uk government links
Transport Secretary Heidi Alexander referred to a "proposed piloting scheme" (Oct 2025); Automated Passenger Services scheme, DfT approval and TfL consent; Automated Vehicles Act 2024 (Sep 2026)
Status now
On 24 Sep 2026 Waymo said it was ready to work with London to earn consent; it gave no launch date.

Labs Round 8 deep profile (31 Oct 2026), as the researchers wrote it; confidence: one source (Waymo only).

Sources

  1. waymo.com/blog/2025/10/hello-london-your-waymo-r
  2. waymo.com/blog/2026/09/visionforlondon/
London; Oxfordwaymo.com/blog/2025/10/hello-london-your-waymo-r · waymo.com/blog/2026/09/visionforlondon/ · techcrunch.com/2026/04/14/london-gets-closer-to- · techcrunch.com/2025/10/15/waymo-plans-to-launch-one source
WayveCompaniesBuilds self-driving software that learns to drive from data and can be fitted to different makes of car.
Profile
Founders
Alex Kendall — Co-founder & CEO (other co-founders not named on pages opened)
Chief executive
Alex Kendall — from null
Chief scientist or research lead
Jamie Shotton — Chief Scientist
Other leaders
Erez Dagan — President; Elisa de Martel — Chief Financial Officer (appointment announced 16 Sep 2026); Emma Baillie — Chief People Officer; Silvius Rus — Sr. Vice President, Engineering; Kaity Fischer — Vice President, Commercial and Operations; Vijay Badrinarayanan — Vice President, AI; Alex Toshev — joined Wayve Labs to lead a new General Robotics Intelligence effort (18 Aug 2026); Andrea Tagliasacchi — Principal Scientist, Wayve Labs (1 Jun 2026)
Hq
London (England: London) — 230-238 York Way, N7 9AG; other_offices: null on pages opened
Headcount
1,200 — 1 July 2026 — https://wayve.ai/thinking/wayve-announces-employee-tender-offer/ (UK headcount not published)
What it does
London company building the "Wayve AI Driver", AI software that learns to drive and can be fitted to different vehicles. It licenses it to carmakers (Mercedes-Benz, Nissan, Stellantis) for supervised assisted driving and to Uber for robotaxis; the first Uber rides using it in the UK began in September 2026.
Research focus
embodied AI for driving; Wayve Labs work on general robotics intelligence and 3D vision (2026 hires)
Models and products
Wayve AI Driver — driving software; production agreement with Mercedes-Benz — 22 Sep 2026; autonomous Uber rides in the UK — 3 Sep 2026; supervised L2+ consumer vehicles planned from 2027 — Feb 2026
Customers or partners named
Mercedes-Benz — production agreement to put the Wayve AI Driver in future vehicles — 22 Sep 2026 — https://wayve.ai/press/wayve-mercedes-benz-announce-production-agreement/; Uber — first autonomous rides in the UK — 3 Sep 2026 — https://wayve.ai/press/category/press-release/; Stellantis — hands-free door-to-door supervised driving — 21 May 2026 — same url; Stellantis and Uber — robotaxis at global scale — 17 Jun 2026 — same url; Stellantis and NVIDIA (with French Government) — autonomous vehicle deployment in France — 23 Oct 2026 — same url; Nissan — automaker investor — 25 Feb 2026 — https://wayve.ai/press/series-d/
Compute
null (Microsoft and NVIDIA are investors; no compute statement found)
Funding total as published
$2.8B "total funding in 4 rounds" — page read 30 Oct 2026 (undated) — https://wayve.ai/company/investors/
Valuation reported
8,600,000,000 USD post-money — 25 Feb 2026 — https://wayve.ai/press/series-d/ and https://wayve.ai/thinking/wayve-announces-employee-tender-offer/ (two company pages; no trade page opened)
Rounds found
2026-02-25 — Series D — 1,200,000,000 USD (headline $1.5B including Uber milestone-based capital) — Eclipse, Balderton, SoftBank Vision Fund 2 (leads) — Microsoft, NVIDIA, Uber, Mercedes-Benz, Nissan, Stellantis, Ontario Teachers' Pension Plan, Baillie Gifford, British Business Bank, Icehouse Ventures, Schroders Capital; 2026-04-15 — Series D extension — 60,000,000 USD — AMD, Arm, Qualcomm Ventures; 2026-07-01 — employee tender (secondary) — 85,000,000 USD — "existing and new investors" (not named)
Uk government links
"Wayve Signs Partnership with UK Government to Accelerate Britain's Self-Driving Future" — 12 May 2026; British Business Bank invested in Series D (Feb 2026)
Status now
Running autonomous rides with Uber in the UK since September 2026 and signed a production agreement with Mercedes-Benz; 1,200 staff as of July 2026.

Labs Round 7 deep profile (30 Oct 2026), as the researchers wrote it; confidence: single_source (company pages only; valuation on two company pages).

Sources

  1. wayve.ai/press/series-d/
  2. wayve.ai/press/series-d/
  3. wayve.ai/press/series-d-extension/
  4. wayve.ai/thinking/wayve-announces-employee-tende
  5. wayve.ai/company/leadership-team/
  6. wayve.ai/press/wayve-mercedes-benz-announce-prod
  7. wayve.ai/press/category/press-release/
  8. wayve.ai/company/investors/
Models and products (13)

Version chainsGAIA-1 (2023) → GAIA-2 (26 Mar 2025) → GAIA-3 (2 Dec 2025)

  1. LA-Pose 2026-05-05 released (paper)A Wayve model that learns vehicle motion from unlabelled driving video to estimate a camera's 3D position.
  2. GAIA-3 2025-12-02 released (research)Wayve's third world model, which generates realistic, controllable driving scenes for offline safety testing.
  3. Rig3R 2025-10-15 released (paper)A Wayve model that builds 3D maps of a scene and tracks the car's movement from several cameras.
  4. GAIA-2 2025-03-26 released (technical report)Wayve's second world model, which generates controllable multi-camera driving video, including rare risky scenarios.
  5. SimLingo 2025-03-12 released (paper)A Wayve research model that drives from cameras alone and links its language to its driving actions.
  6. PRISM-1 2024-06-17 released (internal simulation model)A Wayve model that rebuilds real driving scenes in 3D over time from camera footage, so they can be re-simulated for testing.
  7. WayveScenes101 2024-06-17 released (open dataset, non-commercial)A public set of 101 recorded driving scenes for testing models that render scenes from new viewpoints.
  8. CarLLaVA 2024-06-14 released (paper)A Wayve research model that uses a vision-language model to drive with cameras only.
  9. LINGO-2 2024-04-17 released (tested on public roads)A Wayve driving model that both drives and gives a running commentary explaining its choices.
  10. GAIA-1 2023-10-03 released (technical report)Wayve's first generative world model, which makes realistic driving video from video, text and action inputs to train and test self-driving systems.
  11. LINGO-1 2023-09-14 released (research model)A Wayve model that comments on driving scenes in plain language and answers questions about its driving decisions.
  12. Learning to Drive in a Day (deep reinforcement learning on a car) 2018-09-11 released (paper)Wayve's early research paper in which a real car learned to follow a lane by reinforcement learning.
  13. Wayve AI Driver no date released (product; rides in London with Uber)Wayve's AI driving software for carmakers, from driver assistance up to full self-driving, without HD maps.
230-238 York Way, London (N7 9AG)Companies House · wayve.ai/press/series-d/ · uktech.news/ai/wayve-valuation-massive-series-d- · wayve.ai/press/series-c/ · wayve.ai/press/verified
WeatherNext 3Chips, models and productsGoogle DeepMind's AI global weather model, which produces a new forecast every hour and feeds Google Search, Maps and Gemini. releasedKey facts New forecast every hour; temperature and humidity at 5 km, wind at 10 km; ensemble model taking satellite imagery directly; used by the US National Hurricane Center for Hurricane Melissa's Jamaica landfall; offered via BigQuery, Earth Engine, Google Maps Platform and Cloud Storage. UK scientists offered priority access to WeatherNext (Dec 2025)Made by Google DeepMind · — (no date on page)Londondeepmind.google/science/weathernext/ · deepmind.google/blog/strengthening-our-partnershone source
Wren Dialog AgentChips, models and productsPolyAI agent for designing, building, testing and improving voice and chat agents. live (product)Key facts Listed as a product on PolyAI's technology page and in the May 2026 builder-platform post; documentation and "Start building" link; pricing not statedMade by PolyAI · — (date not stated)Londonpoly.ai/technology · poly.ai/blog/agentic-dialog-platform-open-to-buione source
XLand open-ended learning agentsChips, models and productsDeepMind agents trained across hundreds of thousands of generated 3D games that learned general game-playing behaviour. research (paper only)Key facts Announced 27 Jul 2021; about 700,000 unique games across 4,000 unique worlds; final agents trained on 3.4 million unique tasks; access: paper onlyMade by Google DeepMind · 2021-07-27Londondeepmind.google/blog/generally-capable-agents-emone source

Tile photo: Coal Drops Yard (view from the north) 2025-09-18, Andy Li, CC0 (Wikimedia Commons).