List your company

Mathematical Reasoning on AIME 26 (Accuracy)

74.8AccuracyDynamic Skipping
220.939.858.7May 13, 2026May 14, 2026May 15, 2026May 16, 2026May 17, 2026May 18, 2026May 20, 2026

Evaluation results

MethodLinks
Dynamic SkippingBase Model=GLM-4.7-Flash2026.0574.8
ZEDABase Model=GLM-4.7-Flash2026.0574.4
GLM-4.7-FlashBase Model=GLM-4.7-Flash2026.0574
NETSFT→OPDBase Model=GLM-4.7-Flash2026.0572.9
Dynamic SkippingBase Model=Qwen3-30B-A3B2026.0572.5
ZEDABase Model=Qwen3-30B-A3B2026.0572.5
Qwen3-30B-A3BBase Model=Qwen3-30B-A3B2026.0572.3
NETSFTBase Model=Qwen3-30B-A3B2026.0572.1
ZEDASFTBase Model=GLM-4.7-Flash2026.0571.9
ZEDASFTBase Model=Qwen3-30B-A3B2026.0571.2
NETSFT→OPDBase Model=Qwen3-30B-A3B2026.0570.6
NETSFTBase Model=GLM-4.7-Flash2026.0568.3
AdaMoEBase Model=GLM-4.7-Flash2026.0547.3
AdaMoEBase Model=Qwen3-30B-A3B2026.0536.7
RLVRBackbone=Qwen3-8B-Base, Cost=100%2026.0525.4
RELEXBackbone=Qwen3-8B-Base, Cost=20%2026.0524.6
RLVRBackbone=Qwen3-4B-Base, Cost=100%2026.0523.8
Weight Extrap.Backbone=Qwen3-8B-Base, Cost=20%2026.0523.3
Logits Extrap.Backbone=Qwen3-4B-Base, Cost=15%2026.0520.8
Weight Extrap.Backbone=Qwen3-4B-Base, Cost=15%2026.0520.8
ExPOBackbone=Qwen3-4B-Base, Cost=15%2026.0520
Logits Extrap.Backbone=Qwen3-8B-Base, Cost=20%2026.0520
RELEXBackbone=Qwen3-4B-Base, Cost=15%2026.0519.2
ExPOBackbone=Qwen3-8B-Base, Cost=20%2026.0517.9
AlphaRLBackbone=Qwen3-8B-Base, Cost=20%2026.0517.9
AlphaRLBackbone=Qwen3-4B-Base, Cost=15%2026.0516.3
ConSPOBackbone=Qwen3-4B-Base2026.0512.8
Dr. GRPOBackbone=Qwen3-4B-Base2026.0512.4
GRPOBackbone=Qwen3-4B-Base2026.0510.5
RELEXBackbone=Qwen2.5-Math-1.5B, Cost=15%2026.0510
DisCOBackbone=Qwen3-4B-Base2026.059.4
ExPOBackbone=Qwen2.5-Math-1.5B, Cost=15%2026.058.8
BaseBackbone=Qwen3-4B-Base, Cost=0%2026.058.8
DAPOBackbone=Qwen3-4B-Base2026.058.5
RLVRBackbone=Qwen2.5-Math-1.5B, Cost=100%2026.057.9
Logits Extrap.Backbone=Qwen2.5-Math-1.5B, Cost=15%2026.057.9
Weight Extrap.Backbone=Qwen2.5-Math-1.5B, Cost=15%2026.057.5
BaseBackbone=Qwen3-8B-Base, Cost=0%2026.057.1
AlphaRLBackbone=Qwen2.5-Math-1.5B, Cost=15%2026.055.8
BaseBackbone=Qwen2.5-Math-1.5B, Cost=0%2026.055
Base ModelBackbone=Qwen3-4B-Base2026.054.8