DeepMind alumni's AI agent beats OpenAI's GPT-5.5 in research tasks
What's the story
Inherent, a London-based artificial intelligence (AI) lab founded by former Google DeepMind employees, has announced that its latest AI agent, Faraday, has outperformed larger models from Anthropic and OpenAI. The achievement comes despite Faraday being powered by a much smaller model called Qwen 3.6 with only 27 billion parameters. The task? Independently reproducing the findings of published scientific papers without prior knowledge of the answer.
Performance comparison
Faraday's impressive performance against larger models
When compared to Anthropic's Claude Opus 4.8 and OpenAI's GPT-5.5, both of which are much larger frontier-scale systems, Faraday's performance was impressive.
Inherent co-founder and chief scientist Edward Hughes explained that while beating these advanced agents was nice, what really mattered to them was how they built this capability into Faraday.
The company also wanted its AI agent to show "research taste," an understanding of which experiments are worth running and how to design them well.
Training strategy
Inherent's unique approach to teaching 'taste'
Inherent's approach to teaching "taste" is through reinforcement learning, a method that rewards an AI system for good outcomes instead of dictating rules.
The company hopes this will help its agents generalize better to their long-term goal of creating AI capable of contributing across many scientific fields.
Hughes said, "We're always guided by that north star of building an AI scientist agent and imbuing our agents with taste."
Company culture
Inherent's team and future plans
Inherent's team of 12 works together in an office in London's King's Cross.
Hughes is optimistic about the city's AI talent pool but has also called for an end to "garden leave," a practice that prevents departing employees from joining or starting a rival company for months after they resign.
Despite these challenges, Inherent plans to expand its workforce to about 20-25 by year-end, further establishing its presence in the AI industry.