Metis Reasoning
Category: Foundation Models / LLMs
London AI lab founded by AlphaGo co-creator Thore Graepel building compute-efficient, verifiable machine reasoning systems as an alternative to scaling LLMs. Metis Reasoning was founded in 2026. The company is led by Thore Graepel. Based in London, United Kingdom. Team size: 11-50. Latest round: Undisclosed.
- Founded
- 2026
- Headquarters
- London, United Kingdom
- Team size
- 11-50
Value proposition
AI that produces grounded, auditable and trustworthy conclusions and actions. It separates explicit knowledge representation from the reasoning process, so systems know what they know and don't know, and can be inspected, edited and audited at a fraction of frontier-model compute cost.
Products and solutions
Prototype reasoning system reportedly able to tackle PhD-level scientific problems, an industry-hardened version is expected for external testing within 3-6 months of launch. White papers: 'The Age of Reasoning' and 'The Promise of Principled Reasoning' (Oct 2026).
Unique value
A 'generalisation of AlphaGo': neural perception combined with explicit knowledge representation and structured, hypothesis-testing reasoning. The result is verifiable reasoning traces and guardrails that token-based LLMs cannot provide.
Target customer
Organisations in high-stakes domains where correctness is critical: science, engineering, robotics/embodied AI, medicine and legal reasoning.
Industries served
Robotics / embodied AI, scientific research, engineering, biotech/medicine, legal
Technology advantage
Neuro-symbolic architecture that separates world knowledge from the reasoning process. Verified knowledge is accumulated and reused, which the company says makes it much more compute-efficient than frontier models. Founding team pedigree: AlphaGo/DeepMind, DARPA/Boeing robotics, and Cambridge Consultants.
How they differentiate
Rejects the scaling hypothesis of ever-larger models. Uses explicit knowledge and a structured reasoning architecture with auditable traces instead of chain-of-thought tokens. Targets reliability and safety in high-stakes domains and claims encouraging frontier-reasoning results at a fraction of the cost.
Main competitors
Ineffable Intelligence (David Silver), Recursive Superintelligence (Richard Socher), frontier reasoning-model labs OpenAI, Anthropic and Google DeepMind
Major milestones
Graepel left Google DeepMind in summer 2026. METIS REASONING LIMITED was incorporated at UK Companies House (no. 17419178) on 25 Aug 2026. Bloomberg reported the fundraise in Sep 2026. The company came out of stealth on 19-20 Oct 2026 and published the 'The Age of Reasoning' white paper.
Growth metrics
12-person team at launch (Oct 2026). Bloomberg reported the company is seeking to raise tens of millions of dollars; co-founder Piers Ingram said it has been funded 'more than adequately to date'. The amount is undisclosed.
Market positioning
Alternative-architecture frontier AI lab within London's King's Cross AI cluster, part of the wave of DeepMind-alumni startups pursuing post-LLM approaches.
Geographic focus
UK/Europe (London HQ) with part of the team in Singapore; global ambitions
About Thore Graepel
Ex-Google DeepMind Research Lead and AlphaGo co-creator; previously in AI at Altos Labs; Professor of Computer Science / Chair of Machine Learning at University College London (UCL).
Latest news about Metis Reasoning
More Foundation Models / LLMs companies
Official website: https://metisreasoning.ai/