Mathematical Reasoning on AIME 26 (Accuracy)
74.8AccuracyDynamic Skipping
Evaluation Results
| Method | Links | |
|---|---|---|
| Dynamic SkippingBase Model=GLM-4.7-Flash2026.05 | 74.8 | |
| ZEDABase Model=GLM-4.7-Flash2026.05 | 74.4 | |
| GLM-4.7-FlashBase Model=GLM-4.7-Flash2026.05 | 74 | |
| NETSFT→OPDBase Model=GLM-4.7-Flash2026.05 | 72.9 | |
| Dynamic SkippingBase Model=Qwen3-30B-A3B2026.05 | 72.5 | |
| ZEDABase Model=Qwen3-30B-A3B2026.05 | 72.5 | |
| Qwen3-30B-A3BBase Model=Qwen3-30B-A3B2026.05 | 72.3 | |
| NETSFTBase Model=Qwen3-30B-A3B2026.05 | 72.1 | |
| ZEDASFTBase Model=GLM-4.7-Flash2026.05 | 71.9 | |
| ZEDASFTBase Model=Qwen3-30B-A3B2026.05 | 71.2 | |
| NETSFT→OPDBase Model=Qwen3-30B-A3B2026.05 | 70.6 | |
| NETSFTBase Model=GLM-4.7-Flash2026.05 | 68.3 | |
| AdaMoEBase Model=GLM-4.7-Flash2026.05 | 47.3 | |
| AdaMoEBase Model=Qwen3-30B-A3B2026.05 | 36.7 | |
| RLVRBackbone=Qwen3-8B-Base, Cost=100%2026.05 | 25.4 | |
| RELEXBackbone=Qwen3-8B-Base, Cost=20%2026.05 | 24.6 | |
| RLVRBackbone=Qwen3-4B-Base, Cost=100%2026.05 | 23.8 | |
| Weight Extrap.Backbone=Qwen3-8B-Base, Cost=20%2026.05 | 23.3 | |
| Logits Extrap.Backbone=Qwen3-4B-Base, Cost=15%2026.05 | 20.8 | |
| Weight Extrap.Backbone=Qwen3-4B-Base, Cost=15%2026.05 | 20.8 | |
| ExPOBackbone=Qwen3-4B-Base, Cost=15%2026.05 | 20 | |
| Logits Extrap.Backbone=Qwen3-8B-Base, Cost=20%2026.05 | 20 | |
| RELEXBackbone=Qwen3-4B-Base, Cost=15%2026.05 | 19.2 | |
| ExPOBackbone=Qwen3-8B-Base, Cost=20%2026.05 | 17.9 | |
| AlphaRLBackbone=Qwen3-8B-Base, Cost=20%2026.05 | 17.9 | |
| AlphaRLBackbone=Qwen3-4B-Base, Cost=15%2026.05 | 16.3 | |
| ConSPOBackbone=Qwen3-4B-Base2026.05 | 12.8 | |
| Dr. GRPOBackbone=Qwen3-4B-Base2026.05 | 12.4 | |
| GRPOBackbone=Qwen3-4B-Base2026.05 | 10.5 | |
| RELEXBackbone=Qwen2.5-Math-1.5B, Cost=15%2026.05 | 10 | |
| DisCOBackbone=Qwen3-4B-Base2026.05 | 9.4 | |
| ExPOBackbone=Qwen2.5-Math-1.5B, Cost=15%2026.05 | 8.8 | |
| BaseBackbone=Qwen3-4B-Base, Cost=0%2026.05 | 8.8 | |
| DAPOBackbone=Qwen3-4B-Base2026.05 | 8.5 | |
| RLVRBackbone=Qwen2.5-Math-1.5B, Cost=100%2026.05 | 7.9 | |
| Logits Extrap.Backbone=Qwen2.5-Math-1.5B, Cost=15%2026.05 | 7.9 | |
| Weight Extrap.Backbone=Qwen2.5-Math-1.5B, Cost=15%2026.05 | 7.5 | |
| BaseBackbone=Qwen3-8B-Base, Cost=0%2026.05 | 7.1 | |
| AlphaRLBackbone=Qwen2.5-Math-1.5B, Cost=15%2026.05 | 5.8 | |
| BaseBackbone=Qwen2.5-Math-1.5B, Cost=0%2026.05 | 5 | |
| Base ModelBackbone=Qwen3-4B-Base2026.05 | 4.8 |