Mathematical Reasoning on AIME 25 (Acc (%))
94.4Accuracy (%)GSPO + IBPO(mask)
Evaluation Results
| Method | Links | |
|---|---|---|
| GSPO + IBPO(mask)Model=Qwen3-Next2026.04 | 94.4 | |
| GSPO + IBPOModel=Qwen3-Next2026.04 | 93.8 | |
| GSPO + IBPO(ratio)Model=Qwen3-Next2026.04 | 93.4 | |
| GSPO with Best-of-NModel=Qwen3-Next, N=82026.04 | 91.5 | |
| GSPO with SCoReModel=Qwen3-Next2026.04 | 90.7 | |
| GSPO with promptModel=Qwen3-Next2026.04 | 90.6 | |
| GSPOModel=Qwen3-Next2026.04 | 90.1 | |
| GSPO + IBPO(mask)Model=Qwen3-32B2026.04 | 85.9 | |
| GSPO + IBPOModel=Qwen3-32B2026.04 | 85.3 | |
| GSPO + IBPO(ratio)Model=Qwen3-32B2026.04 | 85 | |
| GSPO with SCoReModel=Qwen3-32B2026.04 | 78.3 | |
| GSPO with promptModel=Qwen3-32B2026.04 | 78.2 | |
| GSPO with Best-of-NModel=Qwen3-32B, N=82026.04 | 77.9 | |
| GSPOModel=Qwen3-32B2026.04 | 77.1 |