Mathematical Problem Solving on MATH-Perturb MP-hard
56.3Exact Match AccuracyDeepMath (RULES)
Evaluation Results
| Method | Links | |
|---|---|---|
| DeepMath (RULES)Optimization=RULES2026.02 | 56.3 | |
| DeepMath (GRPO)Optimization=GRPO2026.02 | 54.5 | |
| DeepMath (vanilla)Optimization=vanilla2026.02 | 53 | |
| Qwen2.5 (RULES)Optimization=RULES2026.02 | 35.1 | |
| Qwen2.5 (vanilla)Optimization=vanilla2026.02 | 33.7 | |
| Qwen2.5 (GRPO)Optimization=GRPO2026.02 | 33.7 | |
| Llama-3.1 (vanilla)Optimization=vanilla2026.02 | 13.3 | |
| Llama-3.1 (GRPO)Optimization=GRPO2026.02 | 13.3 | |
| Llama-3.1 (RULES)Optimization=RULES2026.02 | 12.9 | |
| OLMO2 (RULES)Optimization=RULES2026.02 | 10.4 | |
| OLMO2 (GRPO)Optimization=GRPO2026.02 | 9.3 | |
| OLMO2 (vanilla)Optimization=vanilla2026.02 | 8.6 |