This repository contains three environment variants of Each variant contains the same 4,578 tasks in five Parquet shards, with the original 14-column schema and task order.
Dataset Card
By Loongsage, published under apache-2.0, revision fea0c964cce1.
R2E-Gym: OpenCode, Codex, and Claude Code environments
This repository contains three environment variants of
R2E-Gym/R2E-Gym-Subset.
Each variant contains the same 4,578 tasks in five Parquet shards, with the
original 14-column schema and task order. Only docker_image is replaced with
the corresponding public image containing the selected agent.
| Configuration | Files | Docker image prefix |
|---|---|---|
opencode |
opencode/data/*.parquet |
docker.io/loongsage/r2e-gym:opencode_ |
codex |
codex/data/*.parquet |
docker.io/loongsage/r2e-gym:codex_ |
claude_code |
claude_code/data/*.parquet |
docker.io/loongsage/r2e-gym:claude-code_ |
Images are hosted in loongsage/r2e-gym on Docker Hub. Choose one configuration for a run; combining configurations repeats the same tasks.
from datasets import load_dataset
opencode = load_dataset("loongsage/R2E-Gym", "opencode", split="train")
codex = load_dataset("loongsage/R2E-Gym", "codex", split="train")
claude_code = load_dataset("loongsage/R2E-Gym", "claude_code", split="train")
The task description is problem_statement. The original prompt column is
the meta-prompt used to generate the task, and expected_output_json contains
the original grading information. All original columns are retained, including
commit contents, execution results, modified files and entity summaries.
For Coda, use the r2e_gym data preprocessor and reward, with input_key: prompt
and metadata_key: metadata, and point prompt_data_path at the downloaded
opencode/, codex/, or claude-code/ directory.
The opencode, codex, and claude_code configurations load only their respective folders.
The older root-level prepared-format file is retained separately and is not
included in these configurations.
Upstream dataset license: Apache-2.0.
Structure
opencode 4,578 rows
| Split | Rows | Size |
|---|---|---|
| train | 4,578 | 3.7 GB |
claude_code 4,578 rows
| Split | Rows | Size |
|---|---|---|
| train | 4,578 | 3.7 GB |
codex 4,578 rows
| Split | Rows | Size |
|---|---|---|
| train | 4,578 | 3.7 GB |
Details
- Repository
- loongsage/R2E-Gym
- Publisher
- Loongsage
- Task category
- Text generation
- Tags
- code, r2e-gym
- Size category
- Not stated by the source
- Languages
- en
- Revision
- fea0c964cce1fdab45c3a0ad5e5022d85a010c79
- Last updated
- 2026-09-24
Files
18 files, 797.7 MB in total.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| R2E_Gym_Subset_opencode.parquet | Data | 4.3 MB | 26d698ad9728 |
| claude_code/data/0000.parquet | Data | 55.8 MB | 5651d9046d32 |
| claude_code/data/0001.parquet | Data | 43.4 MB | d00a1e206c5a |
| claude_code/data/0002.parquet | Data | 50.2 MB | d325e15f5913 |
| claude_code/data/0003.parquet | Data | 57.9 MB | c10cb4d763dd |
| claude_code/data/0004.parquet | Data | 57.7 MB | c1c4a4381be9 |
| codex/data/0000.parquet | Data | 55.9 MB | 368913691498 |
| codex/data/0001.parquet | Data | 43.5 MB | c156cdd8a96c |
| codex/data/0002.parquet | Data | 50.0 MB | 84c5686a315e |
| codex/data/0003.parquet | Data | 58.0 MB | 00a849783ae9 |
| codex/data/0004.parquet | Data | 56.8 MB | d6d41d12ce60 |
| opencode/data/0000.parquet | Data | 55.9 MB | 6a277dcd8062 |
| opencode/data/0001.parquet | Data | 43.5 MB | 1d9bc0dd0111 |
| opencode/data/0002.parquet | Data | 50.0 MB | 6c62e4e5d0a2 |
| opencode/data/0003.parquet | Data | 58.0 MB | 1df4bdcf3415 |
| opencode/data/0004.parquet | Data | 56.8 MB | e0c9cba7fdc4 |
| README.md | Documentation | 2.4 KB | — |
| .gitattributes | Repository | 2.5 KB | — |
License and Download
- License
- apache-2.0
- Access
- No access gate
Released by Loongsage through its official repository on Hugging Face. Read the license.