UOJ-Bench - Generation: leaderboard
The code generation task of UOJ-Bench (2026): solve 672 competitive programming problems from IOI-level contests and UOJ original rounds, passing all tests including the extra tests the community added through hacks. UOJ-Bench (2026) draws on real Universal Online Judge submissions and community hacks and judges every answer on the judge itself; Pass@1 in percent, higher is better.
Metric: Pass@1 (%) over all 672 generation problems (UOJ originals and the problems behind the hacking and repair sets, easy to ultrahard): a solution passes when it clears every primary test and every community-added extra test, single-turn zero-shot Pass@1 judged by the Universal Online Judge through its internal API (problem statements translated from Chinese to English; maximum response length at least 64k tokens); higher is better. Source: arxiv.org. Saturation forecast: Around March 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 38.84 | #113 |
| 2 | GPT-5 (High) | 25.6 | #167 (GPT-5) |
| 3 | GPT-5 | 23.96 | #167 |
| 4 | Kimi K2 (Thinking) | 20.68 | #439 (Kimi K2) |
| 5 | DeepSeek V3.2 (Thinking) | 18.75 | #406 (DeepSeek V3.2) |
| 6 | GPT-OSS-120B | 12.35 | #700 |
| 7 | Claude Opus 4.5 | 6.7 | #156 |
| 8 | Qwen 3 Coder 480B A35B Instruct | 2.38 | #643 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=uoj-bench-generation · How It Works · Data refreshed daily, snapshot 2026-11-13.