Inherent, a London AI lab founded by Google DeepMind alumni, says its newly released AI agent has outperformed much larger models from Anthropic and OpenAI at a demanding scientific task: independently reproducing the findings of published research papers without being told the answer in advance.
The claim, reported by TechCrunch on Friday, is the first concrete result from the startup since it emerged from stealth weeks ago with a $50 million seed round. Measured against Anthropic's Claude Opus 4.8 and OpenAI's GPT-5.5 — both frontier-scale systems — Inherent's agent, called Faraday, held its own while running on a comparatively tiny model. For more context on this story, see our ongoing artificial intelligence updates.
A 27-Billion-Parameter Underdog
The most striking detail is the size of the model underneath. Faraday runs on Qwen 3.6, an open-weight model with just 27 billion parameters — a fraction of the size of the frontier systems it was benchmarked against. Parameters are a rough proxy for a model's scale and, typically, its training cost, which makes the comparison a data point in the running debate over whether frontier-scale budgets are necessary for frontier-level behavior on well-scoped tasks.
Inherent framed the benchmark win as a byproduct rather than the goal. "What was most interesting to us about this was not so much the result of beating those frontier agents — which of course we liked — but was actually the way we went about building this," cofounder and chief scientist Edward Hughes told TechCrunch.
The task itself carries real scientific weight. Replicating the findings of published papers, without access to the results, is a standard training exercise for human scientists. "Many PhD students actually start by doing this," Hughes noted.
Teaching an AI 'Research Taste'
Inherent's bar for success was higher than simple accuracy. Beyond replicating results, the company wanted Faraday to demonstrate what it calls research taste — an instinct for which experiments are worth running and how to design them well.
Teaching something as intangible as taste is where reinforcement learning comes in. Rather than spelling out rules for its agents, Inherent rewards good outcomes, betting that the reward-based approach will generalize better to its long-term goal: AI agents capable of contributing 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.
The focus has also shaped what Inherent chose not to build. Rather than developing its own coding tool, the company had Faraday use OpenAI's GPT-5.5 Codex, much as human scientists lean on existing software rather than building everything from scratch.
The team is also trying to avoid building agents that simply tell users what they want to hear. The model, Hughes said, is his favorite kind of teammate — the one who comes back and says: "I got curious about this, and I went off and I did these experiments. What do you think of these results?"
The Team Behind the Bet
Inherent was founded by four researchers with roots across frontier AI. Hughes and cofounder Tantum Collins previously collaborated on cooperative AI research at DeepMind, as did cofounder Louis Kirsch. The fourth cofounder, Kaloyan Aleksiev, came from Reka AI and Microsoft. Collins also brings a policy background rare among AI lab founders, having worked on AI policy in the Biden White House before starting the company. Matt Clifford, cofounder of Entrepreneurs First and the UK government's former AI adviser, has joined as an adviser.
The $50 million seed round was co-led by Index Ventures and Radical Ventures, with participation from Nvidia's venture arm NVentures, Ex/Ante, Metaplanet, Macroscopic Ventures and Mythos Ventures. According to The Next Web, it ranks among Europe's largest AI stealth-to-launch rounds of 2026. Inherent is structured as a public benefit corporation, a legal form that requires it to consider societal impact alongside shareholder returns.
The company is staying deliberately small for now. Its dozen employees all work in person out of an office in King's Cross, the London neighborhood DeepMind helped turn into a global AI hub, and it plans to grow to between 20 and 25 people by the end of the year. Hughes also added his voice to calls to end "garden leave" — the UK practice of barring departing employees from joining or founding a rival for months after resignation — a restriction he said he experienced firsthand and that American researchers generally do not face.
Why a Replication Win Matters
Paper replication occupies an unusual place in science: it is essential to verifying published claims, yet it is slow, unglamorous work that academia rarely rewards. An AI agent that can reliably reproduce published findings — and eventually flag the ones that do not hold up — addresses one of science's most persistent structural problems.
For Inherent, replication is a stepping stone toward a larger ambition the company calls AI-native science: discovery that looks different from the scientific method as practiced for the past 400 years. Index Ventures framed the bet in similar terms when announcing the investment, writing that AI-native science "will be messier, less legible, but capable of exceptional outcomes."
Whether Faraday delivers on that promise will take years to evaluate. But the first result suggests the efficient frontier may be shifting: the ability to direct a small, well-trained model with genuine experimental judgment is emerging as a credible alternative to brute-force scale. For a twelve-person lab in London beating the largest labs in the world at their own evaluation game, that is a strong opening statement.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
