Mathematical Reasoning on Minerva (Avg@16)
46.78Avg@16DISPO
Evaluation Results
| Method | Links | |
|---|---|---|
| DISPOModel=Qwen3-14B-Base2026.02 | 46.78 | |
| CISPOModel=Qwen3-14B-Base2026.02 | 45.66 | |
| DAPOModel=Qwen3-14B-Base2026.02 | 45.22 | |
| DAPOModel=Qwen3-8B-Base2026.02 | 44.85 | |
| DISPOModel=Qwen3-30B-A3B-Base (MoE)2026.02 | 44.6 | |
| DISPOModel=Qwen3-8B-Base2026.02 | 44.6 | |
| CISPOModel=Qwen3-30B-A3B-Base (MoE)2026.02 | 44.26 | |
| CISPOModel=Qwen3-8B-Base2026.02 | 43.52 | |
| DAPOModel=Qwen3-30B-A3B-Base (MoE)2026.02 | 43.45 | |
| Starting checkpointModel=Qwen3-14B-Base2026.02 | 36.1 | |
| S-trace-0.9Backbone=Qwen3-4B2026.05 | 33.28 | |
| DASDModel Scale=8B2026.05 | 33 | |
| DASDModel Scale=4B2026.05 | 31 | |
| Starting checkpointModel=Qwen3-8B-Base2026.02 | 30.95 | |
| GRPO(λ)-0.9Backbone=Qwen3-4B2026.05 | 30.52 | |
| Starting checkpointModel=Qwen3-30B-A3B-Base (MoE)2026.02 | 30.35 | |
| GRPOBackbone=Qwen3-4B2026.05 | 30.02 | |
| RLSDModel Scale=8B2026.05 | 29.7 | |
| HEPOModel Scale=8B2026.05 | 28.7 | |
| OPOBackbone=Qwen3-4B2026.05 | 28.5 | |
| RLSDModel Scale=4B2026.05 | 28.5 | |
| GRPOModel Scale=4B2026.05 | 28.2 | |
| P-trace-0.9Backbone=Qwen3-4B2026.05 | 28.18 | |
| GRPO(λ)-0.99Backbone=Qwen3-4B2026.05 | 27.94 | |
| GRPOModel Scale=8B2026.05 | 27.7 | |
| HEPOModel Scale=4B2026.05 | 27.1 | |
| SRPOModel Scale=4B2026.05 | 26.1 | |
| SRPOModel Scale=8B2026.05 | 25.9 | |
| DASDModel Scale=1.7B2026.05 | 25.3 | |
| RLSDModel Scale=1.7B2026.05 | 25.1 | |
| OPSDModel Scale=4B2026.05 | 24.8 | |
| Qwen3Model Scale=8B2026.05 | 24.7 | |
| Qwen3Model Scale=4B2026.05 | 24.6 | |
| SDPOModel Scale=4B2026.05 | 24.4 | |
| SDPOModel Scale=8B2026.05 | 24.1 | |
| OPSDModel Scale=8B2026.05 | 23.8 | |
| HEPOModel Scale=1.7B2026.05 | 22.8 | |
| GRPOModel Scale=1.7B2026.05 | 22.6 | |
| SRPOModel Scale=1.7B2026.05 | 20.1 | |
| SDPOModel Scale=1.7B2026.05 | 17.8 | |
| Qwen3Model Scale=1.7B2026.05 | 17.5 | |
| OPSDModel Scale=1.7B2026.05 | 17.5 |