| --- |
| pretty_name: DuelLab GameBench 2 results |
| license: cc-by-4.0 |
| language: |
| - en |
| tags: |
| - benchmark |
| - evaluation |
| - code-generation |
| - game-playing |
| - leaderboard |
| configs: |
| - config_name: leaderboard |
| data_files: |
| - split: results |
| path: leaderboard.csv |
| --- |
| |
| # DuelLab GameBench 2 results |
|
|
| This dataset repository is a public mirror of the immutable [DuelLab GameBench 2 release gb2-20260713-16054c869f49](https://benchmarks.duellab.org/releases/gb2-20260713-16054c869f49/). DuelLab is the canonical source; this mirror is generated from its release manifest rather than maintained as a separate leaderboard. |
|
|
| ## Release |
|
|
| - Public release: `gb2-20260713-16054c869f49` |
| - Generated: `2026-07-13T19:49:06Z` |
| - Benchmark version: `2-known8` |
| - Games: 8 |
| - Models: 34 |
| - Reasoning variants: 80 |
| - Rated matches: 56163 |
| - Canonical citation guidance: [duellab.org/citation](https://duellab.org/citation) |
| - Mirror repository: [https://huggingface.co/datasets/DuelLab/gamebench-2-results](https://huggingface.co/datasets/DuelLab/gamebench-2-results) |
|
|
| DuelLab GameBench 2 evaluates AI-generated game-playing programs by compiling and running them head-to-head across an evolving set of named public board and strategy games. Public standings, methodology, and release data are available; the evaluator, engine implementation, and unpublished operational artifacts remain private. |
|
|
| ## Contents |
|
|
| - `leaderboard.csv` and `leaderboard.json`: public leaderboard results. |
| - `entrants.json`: public-safe entrant metadata. |
| - `charts_data.json`: public chart payload. |
| - `benchmark-summary.json`: compact release summary. |
| - `release-manifest.json`: canonical release metadata and artifact hashes. |
| - `citation.*` and `CITATION.cff`: generated citation formats. |
| - `mirror-manifest.json`: hashes for the exact files in this Hub commit. |
|
|
| This is an **evaluation-results dataset**, not a benchmark task corpus and not training data. It does not contain DuelLab's evaluator implementation, private operational artifacts, prompt text, hidden rules, or provider-private fields. GameBench 1 and GameBench 2 are not directly comparable. |
|
|
| ## License and citation |
|
|
| Public release data is licensed under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). Cite the exact immutable DuelLab release using the supplied citation files. Model and provider names remain trademarks of their respective owners. |
|
|
