Inherent, a London AI lab founded by Google DeepMind alumni, says its newly released AI agent, Faraday, outperformed Anthropic鈥檚 Claude Opus 4.8 and OpenAI鈥檚 GPT-5.5 at independently reproducing findings from published scientific papers without being given the answer in advance. Faraday relies on a comparatively small model, Qwen 3.6, which has 27 billion parameters, while its rivals are larger frontier-scale systems. The agent also needed to demonstrate research taste, or an instinct for which experiments are worth running and how to design them, beyond just achieving accurate results.
The startup emerged from stealth weeks ago with a $50 million seed round. Cofounder and chief scientist Edward Hughes said the point was not merely beating frontier agents but the training method behind Faraday. The company uses reinforcement learning, rewarding the AI for good outcomes rather than giving it explicit rules. Inherent is betting this approach will generalize to its longer-term goal of building an AI scientist capable of making new discoveries across many fields.
For now, Faraday did not build its own coding tools. Instead, it used OpenAI鈥檚 GPT-5.5 Codex, similar to how human scientists rely on existing software. Hughes also said the company avoids designing agents that tell users what they want to hear. He described his ideal teammate as one who comes back with unexpected experimental results and asks what the user thinks of them.
Inherent operates with a dozen employees who all work in person at an office in King鈥檚 Cross, a London neighborhood that Google DeepMind鈥檚 presence helped turn into a major AI hub. Hughes said he believes London is the right place for the company. He is also voicing a personal view against garden leave, a U.K. practice that bars departing employees from joining or starting a rival firm for months after resignation, a restriction American researchers typically do not face. Hughes said he was personally affected by that constraint before starting Inherent.
The company was founded by Hughes alongside two other DeepMind alumni and a fourth cofounder. It plans to grow its headcount to roughly 20 to 25 people by the end of the year. With ambitions in world models and recent leadership changes at DeepMind leaving some staff unsettled, Inherent could become an attractive option for DeepMind employees weighing a move.
London鈥檚 AI talent density is a key reason for Inherent鈥檚 location, but the startup also benefits from a U.S. angle: American firms face fewer hiring restrictions than their U.K. counterparts, giving them a head start on recruiting talent. Inherent is not building its own coding tool or training agents primarily on the study of science itself, instead leaning on reward-based learning to teach taste. That focus, Hughes said, guides everything the company chooses to build and what it leaves to existing tools.
More AI news from TechManNews.
