KMP-Bench (Skills) - Similar Problem Generation: leaderboard
Metric: Pass rate (%) of generated similar practice problems and solutions on the final turn of a three-turn generation dialogue, averaged over fail-by-default binary dimensions (construction, solution correctness, solution quality, similarity appropriateness) judged by Gemini-2.0-Flash, 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 | Qwen2.5-Math-72B-Instruct | 82.6 | #708 |
| 2 | Qwen 2.5 72B Instruct | 78.1 | #436 |
| 3 | Qwen 2.5 14B Instruct | 75.9 | #634 |
| 4 | Gemini 2.0 Flash | 73.2 | #331 |
| 5 | Phi-4 | 73.2 | #701 |
| 6 | GPT-4o | 72.2 | #333 |
| 7 | Qwen 2.5 32B Instruct | 71.4 | #491 |
| 8 | Qwen 2.5 7B Instruct | 66.2 | #846 |
| 9 | GPT-4o Mini | 64.8 | #588 |
| 10 | Qwen2.5-Math-7B-Instruct | 52.8 | #1422 |
| 11 | Mistral Nemo Instruct (2407) | 43.9 | #916 |
| 12 | Llama 3.1 8B Instruct | 38.5 | #1018 |
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-similar-problem-generation · How It Works · Data refreshed daily, snapshot 2026-10-11.