Mathematical Reasoning on AIME 2024 (Accuracy, Length)
65.83AccuracyDDPO
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DDPOBackbone=Qwen3-4B2026.03 | 65.83 | 7,972 | |
| DDPOBackbone=Qwen3-8B2026.03 | 65.63 | 7,941 | |
| GRPOBackbone=Qwen3-8B2026.03 | 63.96 | 9,242 | |
| GRPOBackbone=Qwen3-4B2026.03 | 61.87 | 9,288 | |
| DDPOBackbone=Qwen3-14B2026.03 | 60.63 | 8,074 | |
| L1Backbone=Qwen3-14B2026.03 | 59.38 | 8,090 | |
| SHORTER BETTERBackbone=Qwen3-14B2026.03 | 58.96 | 7,246 | |
| A-DLPBackbone=Qwen3-14B2026.03 | 58.75 | 8,519 | |
| GRPO-LEADBackbone=Qwen3-14B2026.03 | 56.88 | 9,010 | |
| GRPOBackbone=Qwen3-14B2026.03 | 56.04 | 8,927 |