Mathematical Problem Solving on MATH-Perturb MP-simple
72.8Exact Match AccuracyDeepMath (GRPO)
Evaluation results
| Method | Links | |
|---|---|---|
| DeepMath (GRPO)Optimization=GRPO2026.02 | 72.8 | |
| DeepMath (RULES)Optimization=RULES2026.02 | 72.8 | |
| DeepMath (vanilla)Optimization=vanilla2026.02 | 72 | |
| Qwen2.5 (RULES)Optimization=RULES2026.02 | 62.4 | |
| Qwen2.5 (vanilla)Optimization=vanilla2026.02 | 59.5 | |
| Qwen2.5 (GRPO)Optimization=GRPO2026.02 | 59.5 | |
| Llama-3.1 (vanilla)Optimization=vanilla2026.02 | 38.7 | |
| Llama-3.1 (RULES)Optimization=RULES2026.02 | 36.9 | |
| Llama-3.1 (GRPO)Optimization=GRPO2026.02 | 36.2 | |
| OLMO2 (RULES)Optimization=RULES2026.02 | 23.3 | |
| OLMO2 (GRPO)Optimization=GRPO2026.02 | 21.9 | |
| OLMO2 (vanilla)Optimization=vanilla2026.02 | 18.6 |