Datasets:

FrontierCS2
/

FrontierCS-v1.1

pretty_name: Frontier-CS 1.1 (FCS1.1) Algorithmic
language:
  - en
license: cc-by-4.0
task_categories:
  - text-generation
size_categories:
  - n<1K
tags:
  - benchmark
  - code
  - algorithms
  - optimization
  - interactive-problems
  - computer-science
configs:
  - config_name: default
    data_files:
      - split: test
        path: data/test.parquet

Frontier-CS 1.1 (FCS1.1): Algorithmic Problems

This is the algorithmic problem dataset for Frontier-CS 1.1 (FCS1.1). It packages 180 algorithmic problems from the Frontier-CS repository, including their original statements, evaluation configurations, checkers or interactors, and released test data. The problems cover challenging algorithm design and optimization tasks, with task-specific evaluation and scoring.

The dataset contains 107 default problems and 73 interactive problems. The source problems are available in the repository's algorithmic/problems directory.

Load the problem table

from datasets import load_dataset

problems = load_dataset("FrontierCS2/FrontierCS-v1.1", split="test")
print(len(problems))  # 180
print(problems[0]["problem_id"])  # "0"
print(problems[0]["statement"])

There is one row per problem, sorted by numeric problem ID. IDs are stored as strings and are not consecutive. The single test split is a packaging choice for benchmark use, not a train/test partition provided by the upstream project. All bundled test files are already public in the source repository.

Data fields

Field	Type	Description
`problem_id`	string	Original problem directory name.
`statement`	string	Complete, unchanged UTF-8 content of `statement.txt`.
`type`	string	`default` or `interactive`, copied from `config.yaml`.
`time_limit`	string	Solution time limit in the source notation, such as `2s`.
`memory_limit`	string	Solution memory limit in the source notation, such as `256m`.
`config_yaml`	string	Complete, unchanged evaluation configuration.
`checker_file`	string or null	Value of the source `checker` key, when present.
`interactor_file`	string or null	Value of the source `interactor` key, when present.
`num_test_inputs`	int64	Number of `testdata/*.in` files.
`num_test_answers`	int64	Number of `testdata/*.ans` files.
`archive_path`	string	`problems.tar.gz`, the single archive containing all 180 problems.
`source_url`	string	Source problem directory URL recorded during packaging.
`source_commit`	string	Source Git commit recorded for data provenance.

The two filename fields expose the original YAML keys without reinterpretation. For evaluation, use the full configuration and the upstream runner.

Full problem files

The single archive problems.tar.gz contains all 180 problems and every tracked file from their source directories, including test data, checkers or interactors, supplied examples, generators, and other auxiliary files. Archive members retain paths under algorithmic/problems/<problem_id>/.

from huggingface_hub import hf_hub_download

archive = hf_hub_download(
    repo_id="FrontierCS2/FrontierCS-v1.1",
    repo_type="dataset",
    filename="problems.tar.gz",
)
print(archive)

To extract all problems with Python 3.12 or later:

import tarfile

with tarfile.open(archive, "r:gz") as bundle:
    bundle.extractall("frontiercs-data", filter="data")

The Parquet table contains statements and metadata. Download problems.tar.gz once to obtain the complete problem bank. Loading the table alone does not download the archive.

README.md
LICENSE
LICENSE-MIT
data/test.parquet
problems.tar.gz
metadata/files.jsonl
metadata/provenance.json

metadata/files.jsonl lists the original path, byte size, and SHA-256 of all 12,299 source files. metadata/provenance.json records packaging provenance, dataset statistics, and archive checksums. The source directories contain 5,861 .in files and 5,861 .ans files; other auxiliary files are also retained.

Evaluation

Use this dataset to browse tasks, construct prompts, and retrieve the released problem assets. Evaluation requires the Frontier-CS evaluation software and its documented runtime dependencies; the archive alone is not a complete judge installation. Follow the installation and evaluation instructions in the upstream repository.

For example, after installing the upstream project and configuring its judge:

frontier eval algorithmic 0 solution.cpp

Interactive problems require the provided interaction protocol and interactor. Scoring is problem-specific: consult each statement, configuration, and checker or interactor. Test answer files are evaluation assets, not a collection of canonical solution programs. Example solutions are included only where they exist in the source directory.

Benchmark use

Because the problems and evaluation assets are public, benchmark results should describe any use of these materials for training, tuning, or iterative feedback. This release does not introduce a hidden evaluation set or new reference scores.

License

The results, patches and traces are released under CC BY 4.0. Code from the upstream research artifacts that appears inside patches or traces stays under its own license.

Citation

@misc{mang2026frontiercs2,
  title     = {{FrontierCS} 2: A Benchmark for Computer Science {R\&D}},
  author    = {Qiuyang Mang and Wenhao Chai and Runyuan He and Dongruixuan Li and Zihan Zheng and
               Yujun He and Xuanyi Zhou and Hao Wang and Siddhant Ray and Bo Peng and Ruyi Ji and
               Shuo Yang and Haocheng Xi and Boru Chen and Xinyang David Han and Melissa Pan and
               Mayank Mishra and Boxi Yu and Sijia Liu and Kerem Akillioglu and Yi Pan and
               Zirui Wang and Mert Cemri and Diogo Antunes and Yichuan Wang and
               Aiwei Ivy Zhang and Zhenyang Chen and Qizheng Zhang and Kaiyuan Liu and
               Wenyuan Huang and Sherry Ruan and Shu Liu and Dacheng Li and Zerui Cheng and
               Zeyu Shen and Shang Zhou and Hanfei Chen and Yuzhong Zhang and Bohan Lyu and
               Qu Ao and Han Zheng and Chonghe Jiang and Jingzhuo Hu and Xiaokun Chen and
               Kunjun Li and Zhiqing Zhong and Haoye Tian and Tong Zhao and Yunzhong He and
               Hanchen Li and Ofir Press and Jiantao Jiao and Kunle Olukotun and
               Karthik R. Narasimhan and Jingbo Shang and Koushik Sen and Matei Zaharia and
               Alex Dimakis and Ion Stoica and Joseph E. Gonzalez and Alvin Cheung},
  year      = {2026},
  url       = {https://frontier-cs.org/}
}

and FrontierCS:

@inproceedings{mang2026frontiercs,
  title     = {{FrontierCS}: Evolving Challenges for Evolving Intelligence},
  author    = {Mang, Qiuyang and Chai, Wenhao and Li, Zhifei and others},
  booktitle = {Proceedings of the International Conference on Machine Learning},
  year      = {2026}
}