Mathematical Problem Solving on AIME 2024 (Pass@k %)
63Pass@kGCPO
Evaluation Results
| Method | Links | |
|---|---|---|
| GCPOk=64, Base Model=Qwen3-1.7B2026.05 | 63 | |
| GCPOk=32, Base Model=Qwen3-1.7B2026.05 | 62.2 | |
| GCPOk=16, Base Model=Qwen3-1.7B2026.05 | 59.1 | |
| DQOk=64, Base Model=Qwen3-1.7B2026.05 | 57.3 | |
| GCPOk=8, Base Model=Qwen3-1.7B2026.05 | 56.3 | |
| GRPOk=64, Base Model=Qwen3-1.7B2026.05 | 54.2 | |
| DQOk=32, Base Model=Qwen3-1.7B2026.05 | 52.8 | |
| GRPOk=32, Base Model=Qwen3-1.7B2026.05 | 49.6 | |
| GCPOk=4, Base Model=Qwen3-1.7B2026.05 | 49.4 | |
| DQOk=16, Base Model=Qwen3-1.7B2026.05 | 47.2 | |
| GRPOk=16, Base Model=Qwen3-1.7B2026.05 | 43.7 | |
| GCPOk=2, Base Model=Qwen3-1.7B2026.05 | 40.7 | |
| DQOk=8, Base Model=Qwen3-1.7B2026.05 | 40.6 | |
| Basek=64, Base Model=Qwen3-1.7B2026.05 | 39.1 | |
| GRPOk=8, Base Model=Qwen3-1.7B2026.05 | 37.4 | |
| Basek=32, Base Model=Qwen3-1.7B2026.05 | 34.5 | |
| DQOk=4, Base Model=Qwen3-1.7B2026.05 | 33.5 | |
| GRPOk=4, Base Model=Qwen3-1.7B2026.05 | 30.2 | |
| Basek=16, Base Model=Qwen3-1.7B2026.05 | 30 | |
| DQOk=2, Base Model=Qwen3-1.7B2026.05 | 28.1 | |
| Basek=8, Base Model=Qwen3-1.7B2026.05 | 25.2 | |
| GRPOk=2, Base Model=Qwen3-1.7B2026.05 | 24.6 | |
| Basek=4, Base Model=Qwen3-1.7B2026.05 | 20.9 | |
| Basek=2, Base Model=Qwen3-1.7B2026.05 | 16.3 |