Ex-DeepMind researcher Thore Graepel launches Metis Reasoning

Summary

Thore Graepel, a former DeepMind researcher who contributed to AlphaGo, has launched Metis Reasoning, a startup focused on developing advanced AI systems with reasoning capabilities. The announcement, reported by Sifted in October 2026, was brief, providing limited information about the company's technical approach, team, or funding.

Why it matters

Reasoning remains a fundamental bottleneck in current large language models (LLMs). The fact that a researcher with an AlphaGo background is establishing a dedicated startup in this area signals that the field is moving beyond pattern matching toward structured inference. This also reflects a broader trend of senior DeepMind/Google AI talent leaving the parent company to work on narrower, high-risk AI capabilities.

Highlights

• Graepel's AlphaGo background lends credibility to the 'reasoning-first' positioning.
• The company is explicitly focused on reasoning rather than general-purpose production.
• Sifted's tags link the startup to AI safety and organizations like Emulate and Ineffable Intelligence, indicating a broader ecosystem.

Critical take

The article is essentially a two-sentence announcement padded with a broken and unreadable block of text. It offers no concrete information regarding funding size, technical architecture, or how it differentiates from existing reasoning efforts (such as o1-style chain-of-thought models). Consequently, the text reads more like a name-drop than a comprehensive news report.