Mathematical Reasoning on AIME 25 (Acc, Think Tok.)
13,040Thinking TokensDRE
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DREBase Model=Qwen3-30B-A3B-Thinking-25072026.06 | 13,040 | 83.54 | |
| DREBase Model=Qwen3-8B2026.06 | 13,448 | 66.67 | |
| DREBase Model=Qwen3-4B-Thinking-25072026.06 | 14,882 | 80.54 | |
| GRPOBase Model=Qwen3-30B-A3B-Thinking-25072026.06 | 15,600 | 81.87 | |
| GRPO + LPBase Model=Qwen3-30B-A3B-Thinking-25072026.06 | 16,206 | 82.71 | |
| GRPOBase Model=Qwen3-8B2026.06 | 16,801 | 64.79 | |
| GRPO + LPBase Model=Qwen3-8B2026.06 | 16,919 | 66.04 | |
| Raw ModelBase Model=Qwen3-8B2026.06 | 17,336 | 68.75 | |
| S-GRPOBase Model=Qwen3-8B2026.06 | 18,090 | 65.83 | |
| Raw ModelBase Model=Qwen3-30B-A3B-Thinking-25072026.06 | 18,095 | 85.62 | |
| S-GRPOBase Model=Qwen3-30B-A3B-Thinking-25072026.06 | 20,501 | 81.67 | |
| GRPOBase Model=Qwen3-4B-Thinking-25072026.06 | 20,930 | 80.83 | |
| GRPO + LPBase Model=Qwen3-4B-Thinking-25072026.06 | 21,059 | 77.08 | |
| Raw ModelBase Model=Qwen3-4B-Thinking-25072026.06 | 21,510 | 79.79 | |
| S-GRPOBase Model=Qwen3-4B-Thinking-25072026.06 | 22,298 | 79.17 |