London-based startup Inherent, founded by Google DeepMind alumni, says its AI agent Faraday has outperformed Anthropic's Claude Opus 4.8 and OpenAI's GPT-5.5 at independently replicating findings from published scientific papers. The company recently emerged from stealth with a $50 million seed round.
Notably, Faraday runs on Qwen 3.6, a comparatively small model with just 27 billion parameters, while the systems it beat are much larger frontier-scale models. Cofounder and chief scientist Edward Hughes said paper replication serves as a standard training exercise for human scientists, with many PhD students starting this way.
Inherent aims to build AI capable of discovering new scientific knowledge rather than just verifying existing results. The company uses reinforcement learning to teach its agents "research taste"—an instinct for what experiments are worth running—rather than primarily training on how science is conducted. Hughes emphasized that the methodology behind achieving the benchmark matters more than the result itself.
The company also chose to have Faraday use OpenAI's GPT-5.5 Codex for coding tasks, mimicking how human scientists rely on existing software rather than building everything themselves.