Inherent's AI agent beats Anthropic, OpenAI at replicating research

Inherent, a London AI lab founded by Google DeepMind alumni, says its AI agent Faraday has outperformed larger, better-known models from Anthropic and OpenAI at a specific task: independently reproducing the findings of published scientific papers without being told the answer in advance. The startup emerged from stealth just weeks ago with a $50 million seed round.

Paper replication is a standard training exercise for human scientists too, cofounder and chief scientist Edward Hughes told TechCrunch, noting that many PhD students start their research careers the same way. Beating other AI systems wasn’t the point, according to Hughes; how Inherent built Faraday was. Measured against Anthropic’s Claude Opus 4.8 and OpenAI’s GPT-5.5 — both much larger, frontier-scale systems — Faraday runs on a comparatively tiny model called Qwen 3.6, with just 27 billion parameters.

Inherent’s bar for success went beyond accuracy: it wanted Faraday to demonstrate “research taste,” an instinct for which experiments are worth running and how to design them well. To teach something that intangible, the company leans on reinforcement learning — rewarding good outcomes rather than spelling out rules — betting the approach will generalize better to its longer-term goal of agents that can contribute across many scientific fields. “We’re always guided by that north star of building an AI scientist agent and imbuing our agents with taste,” Hughes said.

That focus has shaped what Inherent chooses not to build, too: rather than developing its own coding tool, it had Faraday use OpenAI’s GPT-5.5 Codex, much as human scientists lean on existing software rather than building everything themselves. The company is also trying to avoid building agents that simply tell users what they want to hear, aiming instead for a collaborative teammate that comes back with its own findings and questions.

Inherent’s dozen employees all work in person out of an office in King’s Cross, the London neighborhood that Google DeepMind’s presence helped turn into a major AI hub, and the company plans to grow headcount to about 20 to 25 by year’s end. Hughes has also added his voice to calls to end “garden leave,” the UK practice of barring departing employees from joining or starting a rival company for months after resigning — a restriction he says he was personally affected by, and one American researchers generally don’t face.

Why it matters

A tiny agent outperforming frontier-scale models on a scientific task suggests reinforcement-learning-driven “research taste” — not raw parameter count — may be the more defensible edge for AI-for-science startups competing against better-funded labs.