London-based AI lab Inherent has released Faraday, an agent designed to reproduce findings from published scientific papers without being given the answer. The company, founded by Google DeepMind alumni, says the system outperformed agents based on Anthropic’s Claude Opus 4.8 and OpenAI’s GPT-5.5 on this specific task.
Faraday runs on Qwen 3.6, a model with 27 billion parameters, according to the report. Inherent says its evaluation went beyond whether results were replicated: the company also wanted the agent to show “research taste,” including choosing worthwhile experiments and designing them well. It uses reinforcement learning to reward successful outcomes rather than trying to encode every scientific rule directly.
The comparison is a company-reported result on paper replication, not evidence that Faraday can already make original discoveries. Inherent views replication as a training step toward its broader goal of building agents that can contribute new scientific knowledge across fields. The agent also relies on existing tools, using OpenAI’s GPT-5.5 Codex for coding instead of an internally built coding system.
Inherent recently emerged from stealth with a $50 million seed round. Its 12-person team works in person in London’s King’s Cross, and the company plans to expand to roughly 20 to 25 employees by year-end while pursuing AI scientist agents and world models.