Mathematical Reasoning on Olympiad (Pass@k)
49.6Pass@kGCPO
Evaluation Results
| Method | Links | |
|---|---|---|
| GCPOk=64, Backbone=Qwen3-4B2026.05 | 49.6 | |
| GCPOk=32, Backbone=Qwen3-4B2026.05 | 49.1 | |
| DQOk=64, Backbone=Qwen3-4B2026.05 | 48.9 | |
| GCPOk=16, Backbone=Qwen3-4B2026.05 | 48.7 | |
| DQOk=32, Backbone=Qwen3-4B2026.05 | 48.6 | |
| GRPOk=64, Backbone=Qwen3-4B2026.05 | 48.1 | |
| GCPOk=8, Backbone=Qwen3-4B2026.05 | 47.6 | |
| GRPOk=32, Backbone=Qwen3-4B2026.05 | 47.5 | |
| DQOk=16, Backbone=Qwen3-4B2026.05 | 47.2 | |
| Basek=64, Backbone=Qwen3-4B2026.05 | 46.8 | |
| GRPOk=16, Backbone=Qwen3-4B2026.05 | 46.7 | |
| GCPOk=4, Backbone=Qwen3-4B2026.05 | 46.4 | |
| Basek=32, Backbone=Qwen3-4B2026.05 | 46.1 | |
| DQOk=8, Backbone=Qwen3-4B2026.05 | 46 | |
| GRPOk=8, Backbone=Qwen3-4B2026.05 | 45.3 | |
| GCPOk=2, Backbone=Qwen3-4B2026.05 | 44.7 | |
| DQOk=4, Backbone=Qwen3-4B2026.05 | 44.6 | |
| Basek=16, Backbone=Qwen3-4B2026.05 | 44.2 | |
| GRPOk=4, Backbone=Qwen3-4B2026.05 | 43.8 | |
| DQOk=2, Backbone=Qwen3-4B2026.05 | 43.1 | |
| Basek=8, Backbone=Qwen3-4B2026.05 | 42.2 | |
| GRPOk=2, Backbone=Qwen3-4B2026.05 | 41.9 | |
| Basek=4, Backbone=Qwen3-4B2026.05 | 39.7 | |
| Basek=2, Backbone=Qwen3-4B2026.05 | 36.4 |