Faraday AI Agent Shows Remarkable Research Prowess
In a significant development for the AI world, Inherent, a nascent AI lab based in London, has emerged from stealth with an impressive claim: its AI agent, Faraday, has surpassed the performance of larger, more established models from industry giants OpenAI and Anthropic. Founded by former Google DeepMind talent, Inherent is setting its sights on a monumental goal: to develop AI capable of unearthing new scientific knowledge.
Faraday's breakthrough lies in its ability to independently reproduce findings from scientific papers. This critical task involves understanding complex methodologies, experimental setups, and data interpretation to validate published research. In rigorous testing, Faraday successfully replicated results, outperforming both Anthropic's Claude Opus 4.8 and OpenAI's GPT-5.5. What makes this achievement particularly noteworthy is that Faraday accomplished this feat while leveraging a much smaller base model, the Qwen 3.6, with just 27 billion parameters – a fraction of the size of the models it competed against.
This efficiency underscores Inherent's innovative approach and suggests a path toward more resource-light yet powerful AI agents. The company's long-term vision extends beyond replication; they aim to cultivate AI that not only understands but also possesses 'research taste' – an intuitive ability to identify promising avenues for scientific inquiry and drive genuine discovery. Having recently secured a substantial $50 million seed funding round, Inherent is now well-positioned to pursue its ambitious mission to redefine the role of AI in scientific advancement.
