UOJ-Bench - Repair (Easy): leaderboard
Metric: Pass@1 (%) on the 500 easy repair instances (partial-score submissions): the patched code, changing at most 10 percent of the original, must pass every 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 December 2026. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 53.6 |
| 2 | GPT-5 (High) | 43.8 |
| 3 | GPT-5 | 41.4 |
| 4 | Kimi K2 (Thinking) | 32.2 |
| 5 | DeepSeek V3.2 (Thinking) | 26.2 |
| 6 | GPT-OSS-120B | 20.2 |
| 7 | Claude Opus 4.5 | 16.6 |
| 8 | Qwen 3 Coder 480B A35B Instruct | 3.6 |
Interactive version: theaggregate.ai/benchmark?slug=uoj-bench-repair-easy · How It Works · Data refreshed daily, snapshot 2026-09-29.