Inherent, a London AI startup founded by DeepMind alumni, says its Faraday agent outperformed much larger models from Anthropic and OpenAI at replicating scientific research papers.
Faraday runs on a 27 billion parameter model, far smaller than the frontier systems it beat.
The startup raised $50 million and is using reinforcement learning to teach AI systems research judgment, a step toward building AI scientists.
What happened
Inherent, a London-based AI lab founded by Google DeepMind alumni, released Faraday, an AI agent that outperformed Anthropic's Claude Opus 4.8 and OpenAI's GPT-5.5 at independently reproducing published scientific papers. Faraday runs on Qwen 3.6, a model with 27 billion parameters—vastly smaller than the frontier-scale systems it beat.
Why it matters
The startup emerged from stealth just weeks earlier with a $50 million seed round and is testing a novel training approach using reinforcement learning to teach AI systems 'research taste'—an instinct for which experiments are worth running and how to design them well. This method could generalize better to Inherent's longer-term goal of building AI agents capable of discovering new scientific knowledge, not just verifying existing results.
What to watch
Inherent plans to grow from its current dozen employees to about 20 to 25 by the end of the year, all based in its King's Cross office in London. The hiring push may attract departing DeepMind staff, particularly given recent uncertainty at the parent organization.
Ask the AI about this article →
Inherent's announcement arrives at a moment when larger AI labs—OpenAI, Anthropic, Google—dominate headlines with scale-focused announcements. The startup's claim to beat those systems on a specific scientific task using a fraction of their size speaks to an alternative strategy: specialized training methods rather than raw model size. The reinforcement learning approach Hughes describes—rewarding good outcomes rather than prescribing rules—is an explicit bet that this training method will generalize better to the longer-term goal of AI agents that can discover, not just replicate, scientific findings.
The founders' background at DeepMind is significant not just for credibility but for hiring. The article notes that Demis Hassabis's recent role change has left some DeepMind staff unsettled, and Inherent's aggressive hiring plan (targeting 20–25 people by year-end) may benefit from that churn. London's position as an AI hub—driven partly by DeepMind's presence in King's Cross—gives the startup a geographic advantage, though Hughes has criticized U.K. garden leave restrictions that slow hiring compared to U.S. rivals.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
Ask AI anything about this article. Q&As are published on this page for other readers too.
OpenAI Group PBC CEO Sam Altman said in an interview with podcaster David Senra that he is worried AI technolo…
Apple's Siri AI is built on Google's Gemini LLM technology, with Apple Foundation Models refined through knowl…
Cadence Design Systems says rapid changes in AI accelerator power and thermal needs are pushing data-center de…
Marvell Technology disclosed an expanded custom chip agreement with Google on Aug
An anonymous provider has been offering free access to a frontier-class coding model called Ox Alpha on OpenRo…
Hugging Face Inc. is exploring a sale that could value the AI model repository at $13 billion or more, accordi…
