27B Faraday Model Beats Claude Opus at Replicating Research

burny_tech · x · 2026-08-16

Inherent Labs introduced Faraday, a 27B-parameter AI model designed for scientific agents. Trained via long-horizon RL, it outperforms Claude Opus 4.8 and GPT-5.5 in replicating research papers. Key innovations include post-training to use Codex for execution (injecting "taste") and bootstrapping verifiable tasks from existing literature.

Related event: Inherent Labs releases 27B model Faraday for paper reproduction(14 posts)→

Original post →

More from Models

Models channel →