Inherent's 27B Faraday beats larger OpenAI, Anthropic models replicating research

Technology Aug 23, 2026

London-based Inherent, started by former Google DeepMind employees, just revealed its AI agent Faraday beat much larger models from OpenAI and Anthropic at replicating scientific research.

What's wild? Faraday did it with a much smaller model, just 27 billion parameters, showing size isn't everything.

Inherent trains Faraday with 'research taste'

Faraday stands out thanks to Inherent's unique training style that teaches the AI to have "research taste," so it can pick smart experiments on its own.

Instead of reinventing the wheel, it had Faraday use OpenAI's GPT-5.5 Codex instead and focuses on teamwork.

The company is growing its small team in London and recently raised $50 million to push further toward building AIs that can help make real scientific breakthroughs across fields.