Math Reasoning on MATH500 (Pass@1, Pass@k)
68.6Pass@1Qwen3-8B-Base
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Qwen3-8B-BaseTraining=RLAD (ours), Context=30K2026.02 | 68.6 | — | 66.5 | |
| Qwen3-8B-BaseTraining=KDRL, Context=30K2026.02 | 68.4 | — | 64.9 | |
| Qwen3-8B-BaseTraining=GRPO, Context=30K2026.02 | 68.2 | — | 61 | |
| Qwen3-8B-BaseTraining=SFT, Context=30K2026.02 | 67.5 | — | 56 | |
| Qwen3-1.7B-BaseTraining=RLAD (ours), Context=30K2026.02 | 58.1 | — | 39 | |
| Qwen3-1.7B-BaseTraining=KDRL, Context=30K2026.02 | 56.2 | — | 37.2 | |
| Qwen3-1.7B-BaseTraining=GRPO, Context=30K2026.02 | 54.3 | — | 36.5 | |
| IMHModel=WeDLM-8B2026.04 | 54 | 87.5 | — | |
| EntropyModel=WeDLM-8B, Default decoding=true2026.04 | 52.8 | 87.5 | — | |
| ConfidenceModel=WeDLM-8B2026.04 | 52.8 | 85.1 | — | |
| Qwen3-8B-BaseTraining=-, Context=30K2026.02 | 42.1 | — | 36.8 | |
| MarginModel=WeDLM-8B2026.04 | 39.3 | 83.8 | — | |
| IMHModel=LLaDA-8B2026.04 | 36 | 70 | — | |
| ConfidenceModel=LLaDA-8B, Default decoding=true2026.04 | 34.9 | 46.5 | — | |
| Qwen3-1.7B-BaseTraining=-, Context=30K2026.02 | 29.8 | — | 22.2 | |
| EntropyModel=LLaDA-8B2026.04 | 29.4 | 60.6 | — | |
| MarginModel=LLaDA-8B2026.04 | 28.6 | 59.4 | — | |
| RandomModel=LLaDA-8B2026.04 | 25.1 | 63 | — |