UOJ-Bench - Hacking (Easy): leaderboard
Metric: Pass@1 (%) on the 479 easy hacking instances (submissions with overt errors that standard tests expose): the generated input-synthesis script must make the target submission fail on UOJ, 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) | 61.38 |
| 2 | GPT-5 | 51.77 |
| 3 | GPT-5 (High) | 50.73 |
| 4 | Kimi K2 (Thinking) | 45.93 |
| 5 | DeepSeek V3.2 (Thinking) | 40.29 |
| 6 | GPT-OSS-120B | 31.11 |
| 7 | Claude Opus 4.5 | 12.11 |
| 8 | Qwen 3 Coder 480B A35B Instruct | 7.31 |
Interactive version: theaggregate.ai/benchmark?slug=uoj-bench-hacking-easy · How It Works · Data refreshed daily, snapshot 2026-09-29.