---
title: X-Coder-RL-40k
canonical_url: "https://www.modelscope.cn/datasets/IIGroup/X-Coder-RL-40k"
md_url: "https://www.modelscope.cn/datasets/IIGroup/X-Coder-RL-40k.md"
repository: IIGroup/X-Coder-RL-40k
last_updated: 2026-02-07
license: apache-2.0
storage_size: "17 GB"
downloads: 151
stars: 1
---

# X-Coder-RL-40k

> X-Coder-RL-40k - IIGroup 在 ModelScope 开源的数据集。X-Coder-RL-40k is a fully synthetic reinforcement learning dataset for competitive programming, containing 40k high-quality tasks with verified test cases.

IIGroup/X-Coder-RL-40k 是 ModelScope 魔搭社区上的数据集，存储大小 17 GB，采用 apache-2.0 许可。

- **Repository**: IIGroup/X-Coder-RL-40k
- **License**: apache-2.0
- **Storage size**: 17 GB
- **Downloads**: 151
- **Stars**: 1
- **Last updated**: 2026-02-07

Source: https://www.modelscope.cn/datasets/IIGroup/X-Coder-RL-40k

---

# X-Coder-RL-40k

X-Coder-RL-40k is a fully synthetic reinforcement learning dataset for competitive programming, containing 40k high-quality tasks with verified test cases.

## Dataset Description

- **Tasks**: Synthesized by **o3-mini**
- **Test Cases**: Synthesized by **Gemini-2.5-Pro**
- **Total Samples**: ~40k
- **Purpose**: RLVR training for code generation models

## Dataset Structure

The dataset is organized by difficulty level:

| File | Difficulty |
|------|------------|
| `part_0000.parquet` | Easiest |
| `part_0001.parquet` | Easy |
| `part_0002.parquet` | Medium |
| `part_0003.parquet` | Hard |
| `part_0004.parquet` | Hardest |

## Task Difficulty Distribution

**Table: Distribution of Proprietary LLMs' First-Try Pass Rates on Test Cases**

| Pass Rate Range | Qwen3-Max | Gemini-2.5-Pro | GPT5-High |
|-----------------|-----------|----------------|-----------|
| (0–20) | 11.06% | 9.57% | 3.07% |
| [20–40) | 16.44% | 14.38% | 4.83% |
| [40–60) | 18.59% | 17.17% | 6.49% |
| [60–80) | 16.36% | 15.80% | 7.80% |
| [80–100) | 14.39% | 14.90% | 10.82% |
| 100 | 23.16% | 28.18% | 66.98% |

## Usage

```python
from datasets import load_dataset

# Load all data
dataset = load_dataset("IIGroup/X-Coder-RL-40k")

# Load specific difficulty level
easy_tasks = load_dataset("IIGroup/X-Coder-RL-40k", data_files="part_0000.parquet")
hard_tasks = load_dataset("IIGroup/X-Coder-RL-40k", data_files="part_0004.parquet")
```

## Related Resources

- **GitHub**: [X-Coder](https://github.com/JieWu02/X-Coder)
- **SFT Data**: [IIGroup/X-Coder-SFT-376k](https://huggingface.co/datasets/IIGroup/X-Coder-SFT-376k)
- **Models**:
  - [IIGroup/X-Coder-RL-Qwen3-8B](https://huggingface.co/IIGroup/X-Coder-RL-Qwen3-8B)
  - [IIGroup/X-Coder-RL-Qwen2.5-7B](https://huggingface.co/IIGroup/X-Coder-RL-Qwen2.5-7B)

## Citation

```
@article{wu2026x,
  title={X-Coder: Advancing Competitive Programming with Fully Synthetic Tasks, Solutions, and Tests},
  author={Wu, Jie and Li, Haoling and Zhang, Xin and Guo, Jiani and Luo, Jane and Liu, Steven and Huang, Yangyu and Chu, Ruihang and Li, Scarlett and Yang, Yujiu},
  journal={arXiv preprint arXiv:2601.06953},
  year={2026}
}
```

## License

This dataset is licensed under the Apache License 2.0.