LLM Training Puzzles

Toy multi-turn environment where the model must maintain and update an alphabetically sorted list of names across turns; the canonical "hello world" RL env on the Prime Intellect Hub.

Multi-turn open-web research QA environment where the model uses a Serper search tool to answer hard, hop-heavy questions of the BrowseComp / SimpleQA style.

Aligned text-and-3D embodied environment - agents learn household tasks (pick & place, heat, cool, clean) as both TextWorld games and visually-rendered ALFRED scenes.

The canonical Atari 2600 RL benchmark - 57 games (and growing) wrapped in a Gymnasium-compatible interface; the original launching pad for DQN and the entire deep RL field.