KMP-Bench (Dialogue) - Overall Judgement: leaderboard
Metric: Overall judgement accuracy (%): win rate of the judge's holistic decision on the whole response, on KMP-Dialogue: the tutor model continues truncated multi-turn K-8 math tutoring dialogues under a dialogue-specific system prompt naming the target pedagogical principle, and a Gemini-2.0-Flash judge compares each response with the original tutor turn (Win/Tie/Lose); higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.7 Sonnet | 87.1 | #241 |
| 2 | Llama 3.3 70B Instruct | 77.2 | #520 |
| 3 | Qwen 2.5 72B Instruct | 68.5 | #436 |
| 4 | Qwen 2.5 32B Instruct | 67.5 | #491 |
| 5 | Mistral Nemo Instruct (2407) | 66.1 | #916 |
| 6 | Qwen 2.5 14B Instruct | 65 | #634 |
| 7 | Phi-4 | 64 | #701 |
| 8 | Qwen 2.5 7B Instruct | 62.5 | #846 |
| 9 | GPT-4o | 60.7 | #333 |
| 10 | Gemini 2.0 Flash | 58.5 | #331 |
| 11 | Llama 3.1 8B Instruct | 51.3 | #1018 |
| 12 | Qwen2.5-Math-72B-Instruct | 31 | #708 |
| 13 | Qwen2.5-Math-7B-Instruct | 21 | #1422 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=kmp-bench-dialogue-overall-judgement · How It Works · Data refreshed daily, snapshot 2026-10-11.