KMP-Bench (Skills) - Error Diagnosis MR-Score: leaderboard
Metric: MR-Score (%) from MR-GSM8K: 0.2 x MCC of the solution-correctness verdict + 0.3 x first-error-step accuracy + 0.5 x error-reason accuracy, on student solutions with annotated errors, on KMP-Skills (K-8 math problems with LLM-generated pedagogical components); higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 72B Instruct | 65.8 | #436 |
| 2 | GPT-4o | 64.4 | #333 |
| 3 | Qwen 2.5 32B Instruct | 64 | #491 |
| 4 | Phi-4 | 61.4 | #701 |
| 5 | Qwen 2.5 14B Instruct | 58.2 | #634 |
| 6 | GPT-4o Mini | 56.9 | #588 |
| 7 | Qwen2.5-Math-72B-Instruct | 53.8 | #708 |
| 8 | Qwen 2.5 7B Instruct | 34.9 | #846 |
| 9 | Llama 3.1 8B Instruct | 28.6 | #1018 |
| 10 | Mistral Nemo Instruct (2407) | 19.4 | #916 |
| 11 | Qwen2.5-Math-7B-Instruct | 3.4 | #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-skills-error-diagnosis-mr-score · How It Works · Data refreshed daily, snapshot 2026-10-11.