DeepMind alumni startup Inherent says its small AI agent outperformed Anthropic and OpenAI in replicating research

Inherent, a London-based AI lab founded by Google DeepMind alumni, claims its newly released AI agent, Faraday, has outperformed larger models from Anthropic and OpenAI at independently reproducing the findings of published scientific papers, a task that could signal progress toward broader scientific discovery.

What is Faraday and how did it perform?

Faraday, built on a relatively small 27-billion-parameter model called Qwen 3.6, was tested against Anthropic’s Claude Opus 4.8 and OpenAI’s GPT-5.5, both of which are significantly larger frontier-scale systems. Inherent says Faraday achieved higher success in replicating research results without being given the answer in advance. The company emphasized that the goal was not merely to beat these models, but to demonstrate ‘research taste’ — an instinct for designing and prioritizing experiments, which is a more advanced capability than simple pattern matching.

Why this matters for AI research

The ability to independently replicate scientific findings is a fundamental step for any AI system aiming to contribute to scientific discovery. Many PhD students begin their training by replicating existing work, and Inherent co-founder and chief scientist Edward Hughes drew this parallel to highlight the significance of the achievement. The startup’s approach uses reinforcement learning to reward good outcomes rather than explicitly teaching the rules of science, which the team believes will generalize better to their long-term goal of creating AI scientists capable of working across multiple fields.

What sets Inherent apart?

Inherent’s strategy is distinct in that it focuses on a narrow, high-value task rather than trying to build a general-purpose assistant. The company deliberately avoids developing its own coding tools, instead having Faraday use OpenAI’s GPT-5.5 Codex, much like a human scientist would use existing software. This pragmatic approach allows the team to concentrate on what they believe is the core challenge: imbuing AI with ‘taste’ for scientific inquiry.

London as a hub for AI talent

Inherent is headquartered in King’s Cross, a London neighborhood that has become a global AI hub, largely due to Google DeepMind’s presence. The company’s dozen employees work in person, and the founders are bullish on London’s talent density. However, co-founder Edward Hughes has also spoken out against the U.K.’s ‘garden leave’ practice, which he says delayed his own ability to start the company. He argues this restriction puts U.K. startups at a disadvantage compared to U.S. rivals when hiring talent from prior roles.

What’s next for Inherent

Inherent plans to expand its team to about 20-25 employees by the end of the year, and its ambitions extend beyond paper replication to world models and broader AI research. With Demis Hassabis’s new role at Google DeepMind leaving some staff unsettled, Inherent could become an attractive destination for DeepMind researchers considering a move. The company’s progress, though early, suggests that a small, focused team can compete with much larger players in specific scientific tasks.

Conclusion

Inherent’s claim that its small AI agent outperformed frontier models at replicating research is a notable proof point for the effectiveness of focused, reinforcement learning-based approaches. While the company’s ultimate goal of AI-driven scientific discovery remains distant, this early success highlights the potential of nimble, specialized teams to challenge larger incumbents. For now, the scientific community will be watching to see if Faraday’s ‘taste’ translates into genuinely new discoveries.

FAQs

Q1: What is Faraday?
Faraday is an AI agent developed by Inherent, a London-based startup founded by DeepMind alumni. It is designed to autonomously replicate the findings of published scientific papers, a task that requires understanding and executing experimental procedures.

Q2: How does Faraday compare to other AI models?
According to Inherent, Faraday outperformed Anthropic’s Claude Opus 4.8 and OpenAI’s GPT-5.5 in replicating research findings, despite being based on a much smaller model (Qwen 3.6 with 27 billion parameters). The company attributes this to its use of reinforcement learning to train the agent to develop ‘research taste’.

Q3: What is ‘garden leave’ and why is it controversial?
Garden leave is a practice common in the U.K. where departing employees are barred from joining or starting a rival company for a period after resigning. Edward Hughes, Inherent’s co-founder, says this restriction hampers U.K. startups’ ability to hire talent quickly, giving U.S. competitors an advantage.