Mathematical Reasoning on AIME 25 (pass@16)
36.2pass@16P-trace-0.9
Evaluation Results
| Method | Links | |
|---|---|---|
| P-trace-0.9Backbone=Qwen3-4B2026.05 | 36.2 | |
| S-trace-0.9Backbone=Qwen3-4B2026.05 | 32.18 | |
| OPOBackbone=Qwen3-4B2026.05 | 32.06 | |
| GRPOBackbone=Qwen3-4B2026.05 | 26.67 | |
| GRPO(λ)-0.9Backbone=Qwen3-4B2026.05 | 26.66 | |
| GRPO(λ)-0.99Backbone=Qwen3-4B2026.05 | 24.49 |