Mathematical Reasoning on MATH (MV, EM, PK)
89.4Pass@k (PK)Forking
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| ForkingBackbone=Qwen2.5-Math-7B, Method Category=Policy Entropy Method2026.04 | 89.4 | — | — | 84.1 | |
| EAPOBackbone=Qwen2.5-Math-7B, Method Category=Policy Entropy Method2026.04 | 88.7 | — | — | 83.7 | |
| EntroAdvBackbone=Qwen2.5-Math-7B, Method Category=Policy Entropy Method2026.04 | 88.6 | — | — | 83.5 | |
| DAPOBackbone=Qwen2.5-Math-7B, Method Category=Classic RLVR Method2026.04 | 88.1 | — | — | 82.6 | |
| EntroRegBackbone=Qwen2.5-Math-7B, Method Category=Policy Entropy Method2026.04 | 87.7 | — | — | 82.7 | |
| KL-CovBackbone=Qwen2.5-Math-7B, Method Category=Policy Entropy Method2026.04 | 87.2 | — | — | 81.5 | |
| GRPOBackbone=Qwen2.5-Math-7B, Method Category=Classic RLVR Method2026.04 | 85.1 | — | — | 79.9 | |
| W-REINFBackbone=Qwen2.5-Math-7B, Method Category=Classic RLVR Method2026.04 | 84.5 | — | — | 76.2 | |
| I-PPOBackbone=Qwen2.5 3B2026.04 | 79.6 | 66.85 | 51.62 | — | |
| PPOBackbone=Qwen2.5 3B2026.04 | 76.39 | 61.22 | 43.84 | — | |
| PRIMEBackbone=Qwen2.5-Math-7B, Method Category=Classic RLVR Method2026.04 | 74.1 | — | — | 56.6 | |
| p* smooth(alpha = 0.01)Setting=ZVF2026.04 | 73.3 | — | — | — | |
| No samplingSetting=GRPO2026.04 | 73.2 | — | — | — | |
| p*Setting=GRPO2026.04 | 73.2 | — | — | — | |
| No samplingSetting=ZVF2026.04 | 73.2 | — | — | — | |
| p*Setting=ZVF2026.04 | 72.9 | — | — | — | |
| I-PPOBackbone=LLaMa3 8B2026.04 | 72.8 | 55.25 | 43.42 | — | |
| PPOBackbone=LLaMa3 8B2026.04 | 72.77 | 52.08 | 38.71 | — | |
| p* smooth(alpha = 0.01)Setting=GRPO2026.04 | 72.3 | — | — | — | |
| BaselineBackbone=Qwen2.5 3B2026.04 | 71.5 | 51.7 | 35.82 | — | |
| BaselineBackbone=LLaMa3 8B2026.04 | 68.5 | 45.1 | 28.79 | — | |
| I-PPOBackbone=Phi3 4B2026.04 | 67.5 | 52.91 | 46.46 | — | |
| Base-7BBackbone=Qwen2.5-Math-7B, Method Category=Classic RLVR Method2026.04 | 66.3 | — | — | 46.9 | |
| PPOBackbone=Phi3 4B2026.04 | 65.4 | 49.86 | 41.4 | — | |
| BaselineBackbone=Phi3 4B2026.04 | 59.3 | 38.25 | 31.11 | — | |
| I-PPOBackbone=Gemma2 2B2026.04 | 59.1 | 45.07 | 38.71 | — | |
| PPOBackbone=Gemma2 2B2026.04 | 55.56 | 40.66 | 29.36 | — | |
| BaselineBackbone=Gemma2 2B2026.04 | 51 | 33.5 | 20.82 | — | |
| I-PPOBackbone=Rho 1B2026.04 | 49.14 | 23.8 | 19.01 | — | |
| PPOBackbone=Rho 1B2026.04 | 47.62 | 21.4 | 15.33 | — | |
| BaselineBackbone=Rho 1B2026.04 | 41 | 12.2 | 8.41 | — | |
| DeepSeek-V3.2 Exp-Baseshots=4-shot2026.06 | — | — | 60.12 | — | |
| GLM-4.5 Baseshots=4-shot2026.06 | — | — | 57.58 | — | |
| Kimi-K2 Baseshots=4-shot2026.06 | — | — | 68.4 | — | |
| Mistral-Large-3 675B-Base-2512shots=4-shot2026.06 | — | — | 62.88 | — | |
| Nemotron-3-Ultra 550B-A55B-Baseshots=4-shot2026.06 | — | — | 82 | — |