Inherent Releases Faraday, a 27B 'AI Scientist' That Directs a Larger Coding Agent

Inherent, a London AI lab founded by Google DeepMind alumni, released Faraday on August 14, 2026 — a 27B-parameter “AI Scientist” agent¹. Trained via long-horizon reinforcement learning with coding agents as a tool, it outperforms Claude Opus 4.8 and GPT-5.5 on the task of replicating research, the company says¹.

Replica, the RL task space built for training, asks an agent to replicate a figure from a research paper with a limited time and compute budget and without access to the original plot; the initial suite comprises 310 tasks from 100 papers spanning natural language processing, materials science and weather forecasting¹. The notable part is the arrangement: Faraday employs GPT-5.5 Codex as a tool, directing a model several orders of magnitude larger in a way the company says improves replication performance¹. TechCrunch reports that the underlying model is Qwen 3.6 at 27 billion parameters². All the figures come from the company’s own evaluation on its own benchmark, not from third-party verification.

Sources

1. Training AI Scientists to Replicate Research - Inherent Labs official announcement, August 14, 2026
2. Inherent, founded by DeepMind alumni, says its AI ‘teammate’ just outperformed Anthropic and OpenAI at replicating research - TechCrunch, August 22, 2026