Hugging Face launches RL Environment filter with 74 datasets ready for Harbor, Verifiers

NielsRogge · x · 2026-09-26

Hugging Face Datasets now has a dedicated RL Environment filter for reinforcement learning agent training environments, listing 74 datasets that run out of the box with frameworks like Harbor, Verifiers, OpenEnv and NeMo Gym.

Popular entries include terminal-bench-2.1 (177k downloads), terminal-bench (121k), the 30.2M-item procedural-pile, and a science variant of terminal-bench. After community calls for "an HF for RL envs," HF folded the capability into its existing dataset hub — as Niels Rogge quipped, the HF for RL envs is literally just Hugging Face.

Original post →

More from coding & agent

coding & agent channel →