Inherent, an AI lab founded by Google DeepMind alumni, has developed an AI agent called Faraday that outperformed much larger models from Anthropic and OpenAI using significantly less computational resources. The London-based team’s newly released AI agent was able to independently reproduce the findings of published scientific papers without being told the answer in advance.

Faraday ran on a comparatively tiny model called Qwen 3.6, which has just 27 billion parameters, much smaller than Anthropic’s Claude Opus 4.8 and OpenAI’s GPT-5.5, both much larger systems. Inherent aimed for Faraday to demonstrate “research taste,” an instinct for what experiments are worth running and how to design them well. This was achieved through reinforcement learning, a training method that rewards AI systems for good outcomes rather than spelling out rules for them to follow.

Inherent’s approach focuses on building an AI scientist agent imbued with taste, which has shaped their strategy in avoiding the development of certain tools or agents that simply tell users what they want to hear. Instead, Inherent aims for its agents to be collaborative teammates who provide critical feedback and insights based on their own research efforts. The company plans to grow its headcount by the end of the year and is an appealing landing spot for DeepMind employees due to London’s density of AI talent.

This is a summary. Read the original at techcrunch.com.