Ex-Google AI Faraday Beats OpenAI, Anthropic in Research

A London-based startup, Inherent, has developed an AI agent that outperformed models from OpenAI and Anthropic in reproducing scientific research. This agent, named Faraday, was created by ex-Google DeepMind employees. Faraday beat Anthropic's Claude Opus 4.8 and OpenAI's GPT-5.5. What's interesting is that Faraday runs on Qwen 3.6, a much smaller model with only 27 billion parameters. The company trained Faraday using reinforcement learning, a method where the AI is rewarded for achieving good results. This development highlights how smaller, more efficiently trained AI models can compete with larger, more established ones.