JudgeBench Reasoning: leaderboard
JudgeBench reasoning slice evaluates judge-model accuracy on reasoning comparisons from LiveBench-style tasks.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 52 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K2.6 | 96.94 | #104 |
| 2 | Qwen 3.5 35B A3B | 96.94 | #256 |
| 3 | DeepSeek V3.2 Speciale | 96.94 | #128 |
| 4 | Gemini 3.1 Pro (Preview) | 96.43 | #58 |
| 5 | GPT-5.5 | 96.43 | #31 |
| 6 | Qwen 3.5 397B A17B | 95.92 | #146 |
| 7 | GPT-5 Mini | 95.41 | #186 |
| 8 | MiMo-V2-Flash | 95.41 | #372 |
| 9 | Qwen 3.5 9B | 94.9 | #379 |
| 10 | Qwen 3.6 Plus | 94.9 | #120 |
| 11 | Qwen 3.5 122B A10B | 94.9 | #173 |
| 12 | Qwen 3.5 4B | 94.9 | #475 |
| 13 | GLM-5 FP8 | 94.9 | #143 |
| 14 | Kimi K2.5 | 94.39 | #144 |
| 15 | Kimi K2 (Thinking) | 94.39 | #243 (Kimi K2) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=judgebench-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-20.