LiveCodeBench-Plus: leaderboard
LiveCodeBench-Plus (UC Berkeley and Tsinghua IIIS, BenchEvolver paper, 2026) is a 91-problem stdin/stdout competitive-programming set: 64 harder, human-vetted variants evolved from saturated LiveCodeBench-v6 medium and hard problems plus 27 difficult original LiveCodeBench-v6 problems. Pass@1 in percent averaged over four attempts, higher is better; GPT models ran at medium reasoning effort, Gemini models with adaptive reasoning and DeepSeek-V4-Pro in high thinking mode. Items were selected using the failure rates of the models the paper scores, so its ranking is not independent of item selection.
Metric: Pass@1 (%; mean of 4 attempts). Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 (Medium) | 62.6 | #48 (GPT-5.5) |
| 2 | Gemini 3.1 Pro (Preview) | 59.1 | #91 |
| 3 | GPT-5.4 (Medium) | 54.1 | #121 (GPT-5.4) |
| 4 | Gemini 3.5 Flash | 50 | #98 |
| 5 | Qwen 3.7 Max | 47.5 | #142 |
| 6 | Gemini 3 Flash | 40.1 | #159 |
| 7 | GPT-5.4 Mini (Medium) | 29.6 | #389 (GPT-5.4 Mini) |
| 8 | DeepSeek V4 Pro (High) | 27.5 | #181 (DeepSeek V4 Pro) |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=livecodebench-plus · How It Works · Data refreshed daily, snapshot 2026-11-12.