level: business
inherent, a london ai lab founded by google deepmind alumni, released an ai agent called faraday. the company says faraday outperformed larger models from anthropic and openai at independently reproducing findings from published scientific papers. the agent was not given the answers in advance. inherent emerged from stealth weeks ago with a $50 million seed round. the startup aims to build ai that can discover new scientific knowledge, not just verify old results.
faraday runs on qwen 3.6, a model with 27 billion parameters, much smaller than anthropic's claude opus 4.8 or openai's gpt-5.5. inherent measured success not only by accuracy but also by 'research taste', an instinct for designing worthwhile experiments. the company used reinforcement learning, rewarding good outcomes rather than providing explicit rules. faraday used openai's gpt-5.5 codex for coding instead of building its own tool. inherent has about a dozen employees and plans to grow to 20-25 by year end.
paper replication is a standard training exercise for human phd students, according to cofounder edward hughes. inherent's approach focuses on generalization across scientific fields rather than studying how science is conducted. the startup operates from king's cross, london, near google deepmind's offices. hughes has called for ending 'garden leave' restrictions in the uk, which he says slow down startups compared to us competitors. inherent also plans to work on world models.
why it matters: smaller, more efficient ai models could reduce costs and energy use for scientific research automation.