Inherent's Innovative AI Agent Challenges Bigger Models Despite Smaller Scale

Inherent, a London-based AI lab founded by Google DeepMind alumni, has successfully developed an AI agent named Faraday that outperformed larger models from Anthropic and OpenAI in a scientific task.

• Faraday, powered by a comparatively smaller model called Qwen 3.6 with 27 billion parameters, excelled in reproducing scientific paper findings without prior inputs, showcasing its ability against larger models like Anthropic’s Claude Opus 4.8 and OpenAI’s GPT-5.5.
• The AI’s achievement involves demonstrating "research taste," suggesting an instinct for conducting valuable experiments, utilizing reinforcement learning, a method rewarding good outcomes rather than following preset rules.
• Inherent’s founder, Edward Hughes, emphasizes developing AI that collaborates like human scientists, leveraging existing resources and avoiding tools that simply echo user desires.
• Operating from King’s Cross, London, a tech hub partly pioneered by Google DeepMind's presence, Inherent nurtures a collaborative culture and is progressively expanding, aiming to boost its workforce to 20-25 by year-end.
• A point of contention in the UK is the “garden leave” policy that restricts talent movement, which Hughes argues gives US startups an advantage in hiring.

Hughes aspires for Faraday to contribute broadly across scientific fields by continually refining its innovative approach. Despite a quieter launch relative to other DeepMind offshoots, Inherent is planning its expansive move, potentially attracting unsettled DeepMind employees due to organizational changes.