Mathematical Reasoning on AIME 2025 (Pass@k)
73.1Pass@kGCPO
Evaluation Results
| Method | Links | |
|---|---|---|
| GCPOk=64, Backbone=Qwen3-4B2026.05 | 73.1 | |
| GCPOk=32, Backbone=Qwen3-4B2026.05 | 69.8 | |
| DQOk=64, Backbone=Qwen3-4B2026.05 | 67.3 | |
| GCPOk=16, Backbone=Qwen3-4B2026.05 | 66.4 | |
| DQOk=32, Backbone=Qwen3-4B2026.05 | 64.1 | |
| GRPOk=64, Backbone=Qwen3-4B2026.05 | 63.2 | |
| GCPOk=8, Backbone=Qwen3-4B2026.05 | 62.7 | |
| DQOk=16, Backbone=Qwen3-4B2026.05 | 61.2 | |
| GRPOk=32, Backbone=Qwen3-4B2026.05 | 59.9 | |
| GCPOk=4, Backbone=Qwen3-4B2026.05 | 58.9 | |
| GRPOk=16, Backbone=Qwen3-4B2026.05 | 56.9 | |
| DQOk=8, Backbone=Qwen3-4B2026.05 | 55.6 | |
| GCPOk=2, Backbone=Qwen3-4B2026.05 | 53.1 | |
| Basek=64, Backbone=Qwen3-4B2026.05 | 52.7 | |
| GRPOk=8, Backbone=Qwen3-4B2026.05 | 51.7 | |
| DQOk=4, Backbone=Qwen3-4B2026.05 | 51.5 | |
| Basek=32, Backbone=Qwen3-4B2026.05 | 47.8 | |
| GRPOk=4, Backbone=Qwen3-4B2026.05 | 46.8 | |
| DQOk=2, Backbone=Qwen3-4B2026.05 | 44.8 | |
| Basek=16, Backbone=Qwen3-4B2026.05 | 41.4 | |
| GRPOk=2, Backbone=Qwen3-4B2026.05 | 41.2 | |
| Basek=8, Backbone=Qwen3-4B2026.05 | 34.3 | |
| Basek=4, Backbone=Qwen3-4B2026.05 | 27.7 | |
| Basek=2, Backbone=Qwen3-4B2026.05 | 21.8 |