Metis Reasoning, a London-based startup, has officially launched to develop compute-efficient AI models. The company aims to provide an alternative to the large-scale models currently produced by firms like OpenAI and Anthropic, focusing on reasoning capabilities rather than pattern recognition.
The startup is led by co-founder Thore Graepel, a former Google DeepMind research lead and co-creator of AlphaGo. He is joined by Piers Ingram, a serial entrepreneur who has previously raised over $350 million, Michael Warren, an expert in robotics from Boeing, and David Berman, a former AI advisor to NATO and the UK government.
The founding team currently consists of 12 members, including researchers, mathematicians, and robotics specialists. Metis argues that current large language models lack reliability and audit trails, which hinders their adoption in high-stakes fields such as engineering, science, and medicine.
While the company declined to comment on specific funding figures, reports suggest it is raising tens of millions of dollars. The founders state that the company is adequately funded and that their architectural approach is significantly more compute-efficient than existing frontier models.
Metis plans to avoid the industry trend of increasing model sizes and training budgets. The team claims to have already achieved encouraging results in complex reasoning tasks at a fraction of the cost associated with current industry-standard models.
