Inherent, founded by DeepMind alumni, says its AI ‘teammate’ just outperformed Anthropic and OpenAI at replicating research
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Inherent says its Faraday agent outperformed Anthropic's Claude Opus 4.8 and OpenAI's GPT-5.5 at replicating published scientific paper findings — without being told the answers in advance, a task cofounder Edward Hughes likened to work many PhD students start with.
- Faraday runs on Qwen 3.6 with just 27 billion parameters — a fraction of the size of the frontier-scale systems it beat — with Inherent also measuring for "research taste," an instinct for what experiments are worth running.
- Inherent trains agents via reinforcement learning that rewards good outcomes rather than spelling out rules, and had Faraday use OpenAI's GPT-5.5 Codex for coding instead of building its own tool, mimicking how human scientists lean on existing software.
- Inherent emerged from stealth weeks ago with a $50 million seed round and operates as a 12-person team working in person out of a King's Cross office, betting that London's DeepMind-built AI talent density makes it the right base.
- Edward Hughes called for ending UK "garden leave" restrictions, saying the practice — which bars departing employees from joining rivals for months — gives US startups a head start on hiring recently departed talent.
- Inherent plans to grow to 20–25 employees by year-end, and with Demis Hassabis's new role leaving some DeepMind staff unsettled, the startup could become an appealing landing spot for departing DeepMind researchers.
Why it matters: A 12-person London startup with a $50M seed round beating OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.8 at a rigorous scientific task — using a 27-billion-parameter model — suggests capability may depend less on raw scale than on training approach. With DeepMind staff unsettled by Hassabis's new role, Inherent's planned headcount growth positions it to absorb top talent at a fraction of frontier-lab costs.
Ask SkimNews
