DeepMind Alumni's Inherent Says Small AI Agent Beats Frontier Models at Science Replication
Inherent, a London-based AI lab founded by Google DeepMind alumni, says its newly released AI agent, Faraday, outperforms much larger models from Anthropic and OpenAI at independently reproducing the findings of published scientific papers. The claim comes just weeks after the startup emerged from stealth with a $50 million seed round. While better-funded rivals have yet to show concrete results, Inherent is starting to share what it has been building.
Faraday runs on a comparatively tiny model, Qwen 3.6, with just 27 billion parameters, yet measures favorably against Anthropic's Claude Opus 4.8 and OpenAI's GPT-5.5, both frontier-scale systems. Cofounder and chief scientist Edward Hughes says the team cares less about beating rivals and more about how it got there. Inherent also set a higher bar than accuracy alone, training Faraday to demonstrate research taste — an instinct for which experiments are worth running and how to design them well.
To teach that intangible skill, Inherent leans on reinforcement learning, which rewards the AI for good outcomes rather than spelling out explicit rules. The company bets this reward-based approach generalizes better than training on the study of how science is conducted. Paper replication serves as a stepping stone toward Inherent's loftier goal: building AI that can discover new scientific knowledge across many fields, not just verify old results.