Inherent Introduces Faraday AI Scientist Agent

Inherent has introduced Faraday, a 27 billion parameter AI Scientist agent designed to replicate research results. The company stated that Faraday outperformed Claude Opus 4.8 and GPT-5.5 on research replication tasks.

Faraday was trained with long horizon reinforcement learning and uses coding agents as a tool. Inherent said the system directs GPT-5.5 Codex during replication tasks, similar to how researchers use coding assistants.

To train and test Faraday, Inherent developed Replica, a set of reinforcement learning tasks in which an agent must recreate a figure from a research paper within limits on time and compute, without access to the original plot. The first task suite includes 310 tasks from 100 machine learning and AI for science papers across natural language processing, materials science, and weather forecasting.

Inherent said Faraday produced more faithful replications across every category in the suite and had less difficulty with recent research that was not seen by the base model during pretraining. The London based lab recently emerged from stealth with $50 million in seed funding led by Index Ventures and Radical Ventures.