KMP-Bench (Skills) - Error Correction: leaderboard
Metric: Correction accuracy (%): share of turn-two corrected solutions reaching the right answer, 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 | Gemini 2.0 Flash | 96.7 | #331 |
| 2 | Phi-4 | 92.1 | #701 |
| 3 | GPT-4o | 90.9 | #333 |
| 4 | Qwen 2.5 72B Instruct | 89.5 | #436 |
| 5 | Qwen 2.5 32B Instruct | 87.8 | #491 |
| 6 | GPT-4o Mini | 84.3 | #588 |
| 7 | Qwen 2.5 14B Instruct | 82 | #634 |
| 8 | Qwen2.5-Math-72B-Instruct | 81 | #708 |
| 9 | Qwen 2.5 7B Instruct | 73.2 | #846 |
| 10 | Llama 3.1 8B Instruct | 55.9 | #1018 |
| 11 | Qwen2.5-Math-7B-Instruct | 52.3 | #1422 |
| 12 | Mistral Nemo Instruct (2407) | 48.1 | #916 |
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-correction · How It Works · Data refreshed daily, snapshot 2026-10-11.