JudgeBench Math: leaderboard
JudgeBench math slice evaluates judge-model accuracy on mathematical answer comparisons.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 52 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.6 Plus | 96.43 | #120 |
| 2 | Qwen 3.5 397B A17B | 94.64 | #146 |
| 3 | Qwen 3.6 35B A3B | 94.64 | #226 |
| 4 | Kimi K2.5 | 93.75 | #144 |
| 5 | Qwen 3.5 9B | 93.75 | #379 |
| 6 | Qwen 3.5 4B | 93.75 | #475 |
| 7 | GPT-5 Mini | 92.86 | #186 |
| 8 | Qwen 3.5 35B A3B | 92.86 | #256 |
| 9 | MiniMax-M2.5 | 92.86 | #303 |
| 10 | MiMo-V2-Flash | 92.86 | #372 |
| 11 | GLM-5 FP8 | 92.86 | #143 |
| 12 | Gemini 3.1 Pro (Preview) | 91.96 | #58 |
| 13 | Qwen 3.5 122B A10B | 91.96 | #173 |
| 14 | DeepSeek V3.2 Speciale | 91.96 | #128 |
| 15 | GPT-5.5 | 91.07 | #31 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=judgebench-math · How It Works · Data refreshed daily, snapshot 2026-10-20.