Inherent, a London AI lab founded by former Google DeepMind staff, said its newly released agent Faraday has independently reproduced findings from published scientific papers and outperformed larger systems on that task. The company reported the comparison included Anthropic’s Claude Opus 4.8 and OpenAI’s GPT-5.5.
Size and experimental result
Inherent said Faraday runs on Qwen 3.6, a model with about 27 billion parameters, a fraction of the size of the frontier systems it was measured against. The startup reported the evaluation aimed not only for accuracy in replication but also for an agent-level sense of which experiments were worth pursuing—what the team calls “research taste.”
Co-founder and chief scientist Edward Hughes said replication is a common training exercise for human researchers: “Many PhD students actually start by doing this.” Hughes added that the team was more interested in the training approach than in merely beating other agents.
Training approach and tooling choices
Inherent described its method as leaning on reinforcement learning, which rewards desirable outcomes instead of prescribing step-by-step rules. The company said it prefers this reward-based path because it hopes that it will generalize to agents capable of contributing across multiple scientific fields.
The team also chose not to build a bespoke coding tool for Faraday. Instead, Inherent said Faraday used OpenAI’s GPT-5.5 Codex for code generation and execution, likening the choice to how human scientists reuse existing software rather than rebuilding it from scratch.
Inherent emerged from stealth weeks earlier with a $50 million seed round, the company said. The startup reported its roughly dozen employees work together in an office in King’s Cross, London, and plans to expand to “about 20 to 25” staff by the end of the year.
Hughes also voiced a personal view on U.K. hiring rules, saying he supports ending “garden leave,” a practice that can delay employees from joining rival firms after departing a role. He framed that comment as personal rather than a company position.
Original source: TechCrunch AI