🏆 ALE-Bench Leaderboard
-E10600?logo=data:image/png;base64,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){kind=icon}
-E10600?logo=data:image/png;base64,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){kind=icon}
ALE-Bench is a benchmark for evaluating AI systems on score-based algorithmic programming contests. Drawing on real-world tasks from the AtCoder Heuristic Contest (AHC), ALE-Bench presents optimization problems (e.g., routing and scheduling) that are computationally hard and admit no known exact solution.
This page displays the leaderboard for ALE-Bench, showcasing the performance of various AI models on the benchmark tasks.
Current version is updated as of 2026-09-16. For the initial version of the leaderboard (as of 2025-06-17), please visit here.
All JSON result data published on this webpage can be downloaded in bulk with download_results_json.sh.
Overall Leaderboard
Experimental Setup: Self-refine x1 involved 15 sampling iterations, from which the response getting the median score across 50 local cases was selected. Details can be found in the ALE-Bench GitHub repository ( LLMs' setting and implementation code ).
Benchmark Results by Problem
Benchmark Results by Release Date
AI Rankings Across Problems
Compare models by their mean AI rank across all problems.
How the average ranks are calculated
We average the AI ranks from the problem tables below, using the same configuration selection for every problem. Tied scores share competition ranks such as 1, 1, 3. The human reference contributes to relative scoring but has no AI rank and is excluded from this average.
All entries must cover every problem. Each problem has equal weight regardless of its case count. Mean rank measures relative position, so it does not capture the size of score differences. Adding models can change the ranks and their averages.
Cost and token values are means over the same problems. Hiding a model in the legend only changes visibility and does not recalculate ranks.
AI Rankings by Problem
Compare models on the same private cases. Click a model to view its code.
| Self-refine |
|---|
How these rankings are calculated
Ranking method:
The human reference includes only submissions made during the contest. Better scores achieved on the post-contest overtime leaderboard are not included.
Higher total scores rank first.
The human reference combines the best score for each case across multiple participants. It does not represent a single person's submission and may not be achievable with a single program.
With relative scoring, adding a new model can change the scores and rankings of existing entries. The scores shown in the Run Summary section below may differ because these rankings use case-wise best scores or ranks across the selected AI entries and the human reference, not because of a bug in the scoring method.
Run Summary and Code