Mathematical Reasoning on AMC (Accuracy, Length)
91.56AccuracyDDPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DDPOBackbone=Qwen3-4B2026.03 | 91.56 | 4,074 | |
| DDPOBackbone=Qwen3-14B2026.03 | 91.09 | 3,622 | |
| DDPOBackbone=Qwen3-8B2026.03 | 90.47 | 4,445 | |
| L1Backbone=Qwen3-14B2026.03 | 90.31 | 4,957 | |
| GRPOBackbone=Qwen3-8B2026.03 | 90.31 | 6,013 | |
| GRPOBackbone=Qwen3-14B2026.03 | 89.84 | 4,080 | |
| SHORTER BETTERBackbone=Qwen3-14B2026.03 | 88.44 | 3,600 | |
| GRPO-LEADBackbone=Qwen3-14B2026.03 | 88.44 | 4,823 | |
| A-DLPBackbone=Qwen3-14B2026.03 | 88.13 | 3,998 | |
| GRPOBackbone=Qwen3-4B2026.03 | 88.13 | 6,031 |