Key facts
- Inherent, an AI startup founded by former Google DeepMind employees, has launched its AI agent named Faraday.
- Faraday reportedly outperformed Anthropic's Claude Opus 4.8 and OpenAI's GPT-5.5 in replicating scientific research findings.
- Faraday operates on a significantly smaller model, Qwen 3.6 with 27 billion parameters, compared to its larger competitors.
- The startup's goal is to develop AI agents that can discover new scientific knowledge and possess 'research taste'.
- Inherent is based in London and plans to expand its team to 20-25 employees by year-end.
Inherent, an AI startup founded by alumni of Google DeepMind, announced that its AI agent, Faraday, has successfully replicated findings from published scientific papers, outperforming larger models from Anthropic and OpenAI. The London-based company, which recently secured $50 million in seed funding, highlighted that Faraday operates on a comparatively small model, Qwen 3.6 with 27 billion parameters, while demonstrating 'research taste'—an instinct for valuable experiments and their design.
Co-founder and chief scientist Edward Hughes stated that while beating frontier agents was positive, the method of building Faraday was more significant. He explained that paper replication is a standard training exercise for human scientists. Inherent's approach leverages reinforcement learning, rewarding good outcomes rather than explicitly defining rules, to foster generalization for its long-term goal of creating AI scientists capable of discovering new knowledge.
Faraday utilized OpenAI's GPT-5.5 Codex for coding tasks, mirroring how human scientists use existing software. Hughes described the ideal AI teammate as one that proactively conducts experiments and shares results. The company, comprising a dozen employees working in person in London, plans to expand its team to 20-25 people by the end of the year. Hughes also voiced support for ending 'garden leave' practices in the UK to better compete for talent.