Mathematical Reasoning on AIME 2025 (Acc avg@32)
93.2Accuracy (avg@32)DFPO (Min-Replace)
Evaluation Results
| Method | Links | |
|---|---|---|
| DFPO (Min-Replace)Backbone=Qwen3-Next-80B-A3B-Thinking2026.04 | 93.2 | |
| DFPO (Orth-Proj)Backbone=Qwen3-Next-80B-A3B-Thinking2026.04 | 93.1 | |
| GRPO-fixBackbone=Qwen3-Next-80B-A3B-Thinking2026.04 | 91.9 | |
| GSPOBackbone=Qwen3-Next-80B-A3B-Thinking2026.04 | 89.8 | |
| GRPOBackbone=Qwen3-Next-80B-A3B-Thinking2026.04 | 89.7 | |
| DFPO (Orth-Proj)Backbone=Qwen3-32B2026.04 | 82.6 | |
| DFPO (Min-Replace)Backbone=Qwen3-32B2026.04 | 82.5 | |
| GRPO-fixBackbone=Qwen3-32B2026.04 | 80.6 | |
| GSPOBackbone=Qwen3-32B2026.04 | 76.9 | |
| GRPOBackbone=Qwen3-32B2026.04 | 76.9 |