README.md

---
license: cc-by-4.0
---

# LiveOIBench Tests

LiveOIBnehch consists of 403 coding problems collected directly from the official websites of 72 competitions across 14 renowned Informatics Olympiads, focusing on contests held from 2023 onward. We collect all the offical test cases, human contestant ranking results, and contestant Codeforces profiles.

Please refer to the [paper](https://arxiv.org/abs/2510.09595) for more details and the [leaderboard](https://liveoibench.github.io/leaderboard.html) for the most up-to-date models' performance.

## Usage

Please see this [Github repository](https://github.com/LiveOIBench/LiveOIBench-Evaluation) for downloading and processing the dataset.

## Files

- `liveoibench_testcases_v1_2023.parquet` – testcases for problems from 2023.
- `liveoibench_testcases_v1_2024.parquet` – testcases for problems from 2024.
- `liveoibench_testcases_v1_2025.parquet` – testcases for problems from 2025/01 to 2025/05.

The test files are large; please download them with caution.

## Schema

Each row in the testcases Parquet files has the following fields:

- `id`: Unique integer identifier for the row.
- `problem_id`: Identifier of the corresponding problem, matching
  the `problem_id` column in LiveOIBench.
- `subtasks`: JSON‑encoded mapping from subtask indices to
  subtask metadata. Each entry typically includes:
  - `score` (numeric subtask score),
  - `testcases` (list of testcase identifiers assigned to the subtask), and
  - `task` (optional descriptive label for the subtask).
- `tests`: JSON‑encoded mapping from testcase identifiers to
  their I/O content. For each testcase id, the value is an object containing
  at least:
  - `input` (raw input string passed to the solution),
  - `output` (expected output string from the official judge).

Input and output strings are stored exactly as used by the official judge,
including newlines and any non‑ASCII characters.

## License

This project is licensed under CC-BY-4.0. Please respect the original contest rules and attribution when using the
problems.

## Citation

If you find LiveOIBench and the repo useful in your work, please cite:

```bibtex
@article{zou2025liveoibench,
  title={LiveOIBench: Can Large Language Models Outperform Human Contestants in Informatics Olympiads?},
  author={Zou, Kaijian and Xiong, Aaron and Zhang, Yunxiang and Zhang, Frederick and Ren, Yueqi and Yang, Jirong and Lee, Ayoung and Bhushan, Shitanshu and Wang, Lu},
  journal={arXiv preprint arXiv:2510.09595},
  year={2025},
  url={https://arxiv.org/abs/2510.09595},
  doi={10.48550/arXiv.2510.09595}
}
```