List your company

Mathematical Problem Solving on MATH-Perturb MP-hard

56.3Exact Match AccuracyDeepMath (RULES)
6.69219.57132.4545.329Feb 1, 2026

Evaluation results

MethodLinks
DeepMath (RULES)Optimization=RULES2026.0256.3
DeepMath (GRPO)Optimization=GRPO2026.0254.5
DeepMath (vanilla)Optimization=vanilla2026.0253
Qwen2.5 (RULES)Optimization=RULES2026.0235.1
Qwen2.5 (vanilla)Optimization=vanilla2026.0233.7
Qwen2.5 (GRPO)Optimization=GRPO2026.0233.7
Llama-3.1 (vanilla)Optimization=vanilla2026.0213.3
Llama-3.1 (GRPO)Optimization=GRPO2026.0213.3
Llama-3.1 (RULES)Optimization=RULES2026.0212.9
OLMO2 (RULES)Optimization=RULES2026.0210.4
OLMO2 (GRPO)Optimization=GRPO2026.029.3
OLMO2 (vanilla)Optimization=vanilla2026.028.6