Faraday AI Beats GPT-5.5 in Research Tasks: DeepMind Alumni

An AI agent called Faraday has surpassed larger models from Anthropic and OpenAI in research tasks. Faraday, developed by Inherent, a London-based AI lab, achieved this despite being powered by a much smaller model with only 27 billion parameters. What's interesting is that Faraday independently reproduced findings from published scientific papers without prior knowledge of the answers. This performance was compared against models like Anthropic's Claude Opus 4.8 and OpenAI's GPT-5.5. Inherent's co-founder noted that while beating these advanced agents is good, the key is how they built this capability. They aim for Faraday to demonstrate "research taste," understanding which experiments are valuable and how to design them. The company uses reinforcement learning to teach this "taste," rewarding good outcomes. This approach could help AI agents contribute across many scientific fields. This development shows how smaller, focused AI models can achieve significant breakthroughs.