Dataset catalog

External · RLE-1204

Terminal-Bench

Benchmark of 89 hard, realistic command-line tasks run in Docker; agents issue tmux/bash keystrokes verified by outcome tests.

Type RL environmentsVolume 89 tasksFormat Docker + YAML/dir task specs + Python harnessAccess PUBLIC LICENSE

External · RLE-1403

Reasoning Gym

Python library of 100+ procedural dataset generators with algorithmic verifiers for RL with verifiable rewards; generates virtually unlimited reasoning problems with adjustable difficulty.

Type RL environmentsVolume 100+ generatorsFormat Python (pip: reasoning-gym)Access PUBLIC LICENSE

External · RLE-1404

SkyRL-Gym

Gymnasium-API library of tool-use environments (math, code, search, text-to-SQL) for LLM post-training, part of the SkyRL RL stack from NovaSky.

Type RL environmentsVolume tool-use environment library (math/code/search/SQL)Format Python (Gymnasium API; pip)Access PUBLIC LICENSE

External · RLE-1305

AgentGym (14 environments)

Unified framework of 14 interactive environments across 7 scenario types with a common HTTP/ReAct interface, plus trajectory datasets and the AgentEval benchmark.

Type RL environmentsVolume 14 environmentsFormat Python + HTTP env servers (agentenv)Access PUBLIC LICENSE

External · SFT-2310

Nemotron-Post-Training-Dataset-v1

NVIDIA 25.6M-row post-training corpus (chat/code/math/stem/tool_calling) — includes a rare 310K genuine tool-calling split with real tool_calls schemas; CC-BY-4.0.

Type SFT datasetVolume 25659642 rowsFormat Parquet (Hugging Face)Access PUBLIC LICENSE

External · RLE-1207

tau2-bench (τ²-bench)

Dual-control tool-agent benchmark (278 tasks: retail 114, telecom 114, airline 50) where both agent and user can call tools.

Type RL environmentsVolume 278 tasks (retail 114 + telecom 114 + airline 50)Format Python package + JSON domain dataAccess PUBLIC LICENSE

External · SFT-2307

OpenThoughts-114k

114K verified DeepSeek-R1 reasoning traces over math, science, code and puzzles; Open Thoughts, Apache-2.0.

Type SFT datasetVolume 114000 rowsFormat Parquet (Hugging Face)Access PUBLIC LICENSE

Type RL environmentsVolume retrieval-augmented QA training frameworkFormat Python (built on veRL; installed from source via conda/pip)Access PUBLIC LICENSE

External · RLE-1402

Search-R1

Open-source RL framework that trains LLMs to interleave reasoning with live search-engine calls; the retriever is treated as part of the RL environment.

External · SFT-2205

OpenMathReasoning

~5.68M math solutions over 306k unique AoPS problems, split into CoT, tool-integrated reasoning (Python code) and GenSelect.

Type SFT datasetVolume ~5.68M solutionsFormat Parquet (Hugging Face)Access PUBLIC LICENSE

External · RLE-1401

RAGEN Environments

Reinforcement-learning framework with 10 stylized interactive environments (Sokoban, FrozenLake, Bandit, Countdown, Sudoku, WebShop, etc.) for training multi-turn reasoning agents.

Type RL environmentsVolume 10 environmentsFormat Python (Gym-compatible interface; installed from source via setup script)Access PUBLIC LICENSE

External · RLE-1306

TextArena

Open collection of 100+ competitive/cooperative text games with an OpenAI-Gym-style interface, online play, and a TrueSkill leaderboard for LLM agents.

Type RL environmentsVolume 100+ gamesFormat Python (pip: textarena) + Gym-style APIAccess PUBLIC LICENSE