Inherent Claims AI Agent Outperforms Giants in Science Replication
AI-generated image · US National Wire
Founded by Google DeepMind alumni, the London startup says its Faraday agent can reproduce scientific findings using a fraction of the parameters of frontier models.
Inherent, a London-based AI lab founded by former Google DeepMind employees, claims its new AI agent, Faraday, has outperformed larger models from OpenAI and Anthropic at reproducing the findings of published scientific papers, as TechCrunch first reported. According to that reporting, the startup achieved this result using Qwen 3.6, a model with 27 billion parameters, which is significantly smaller than OpenAI’s GPT-5.5 and Anthropic’s Claude Opus 4.8.
While the company's ultimate goal is to build AI capable of discovering new scientific knowledge, cofounder and chief scientist Edward Hughes told TechCrunch that replicating papers is a foundational exercise for scientists. To achieve this, Inherent utilized reinforcement learning to imbue Faraday with "research taste"—an instinct for designing and running valuable experiments—rather than relying on rules or the study of scientific conduct.
Faraday does not operate entirely independently; Inherent integrated OpenAI’s GPT-5.5 Codex to handle coding tasks. The startup, which recently emerged from stealth with a $50 million seed round, is currently staffed by a dozen employees working in person in King’s Cross, London. Co-founders Louis Kirsch, Kaloyan Aleksiev, Tantum Collins, and Edward Hughes plan to expand the team to between 20 and 25 people by the end of the year.