Clem Delangue: open RL environments are the new open pretraining data

Thom_Wolf · x · 2026-09-22

HuggingFace CEO Clem Delangue praised a team that shipped a top-6 Artificial Analysis model plus tech report less than a week after starting its final RL run, releasing 7k RL training samples and the framework. He argues releasing high-quality open-source RL environments is the most impactful thing anyone can do for the open frontier—the RLVR-era equivalent of sharing quality pretraining data.

Original post →

More from AGI Musings

AGI Musings channel →