British AI lab Inherent, founded by DeepMind alumni, launched Faraday on August 22 — an agent the company positions as an AI research "teammate" rather than a chat assistant. Its headline claim: Faraday beat Anthropic's and OpenAI's models at replicating scientific papers.
Replication is a deliberate choice of target. Taking a published paper and reproducing its results end to end is a task with a ground truth, which makes it one of the few agentic science benchmarks that can be scored at all — and Inherent frames it as the stepping stone toward agents that produce novel results rather than re-derive known ones.
The comparison is Inherent's own, reported by TechCrunch, with no independent evaluation alongside it. Treat a startup's claim to have beaten both frontier labs on a self-selected task as a marketing artifact until someone else runs the numbers.
Sources: TechCrunch
