Inherent releases Faraday, an AI tool for replicating science papers

In short: Inherent, a London AI lab founded by former DeepMind researchers, released an AI agent called Faraday that it says is good at redoing published scientific studies and checking if the results hold up.

What happened

Inherent says Faraday can independently reproduce the findings of scientific papers without being told the answer ahead of time. Think of it like a lab assistant that reads a recipe, then tries to cook the dish on its own to see if it really works.

The company told TechCrunch that Faraday performed better at this specific task than larger AI systems from OpenAI and Anthropic. Those include OpenAI’s GPT-5.5 and Anthropic’s Claude Opus 4.8, according to the report.

One reason Inherent highlights is that Faraday runs on a smaller underlying model called Qwen 3.6, which has 27 billion “parameters” (parameters are like the number of adjustable knobs in the system, and more knobs usually means higher training costs). Inherent also said it aimed for more than just correct answers. It wanted Faraday to show “research taste,” meaning it can pick sensible experiments to run and plan them well.

To train this, Inherent used reinforcement learning, which is a way of training by giving rewards for good outcomes instead of writing step by step rules (like training a dog with treats rather than giving a long instruction manual). The company also said it had Faraday use OpenAI’s GPT-5.5 Codex for coding tasks, rather than building its own coding tool.

Why it matters

If tools like this work reliably, they could help researchers double-check results faster and catch mistakes earlier. That can save time and money, and it can make it easier to trust what gets published.

Source: TechCrunch AI