List your company

Mathematical Reasoning on HMMT25 (accuracy)

16.3Accuracy (HMMT25)RLVR
-0.6523.7498.1512.551May 20, 2026

Evaluation results

MethodLinks
RLVRBackbone=Qwen3-8B-Base, Cost=100%2026.0516.3
Weight Extrap.Backbone=Qwen3-8B-Base, Cost=20%2026.0516.3
RELEXBackbone=Qwen3-8B-Base, Cost=20%2026.0515.4
RELEXBackbone=Qwen3-4B-Base, Cost=15%2026.0514.6
ExPOBackbone=Qwen3-8B-Base, Cost=20%2026.0513.3
AlphaRLBackbone=Qwen3-8B-Base, Cost=20%2026.0511.3
Weight Extrap.Backbone=Qwen3-4B-Base, Cost=15%2026.0510.4
RLVRBackbone=Qwen3-4B-Base, Cost=100%2026.0510
ExPOBackbone=Qwen3-4B-Base, Cost=15%2026.059.6
Logits Extrap.Backbone=Qwen3-8B-Base, Cost=20%2026.058.3
AlphaRLBackbone=Qwen3-4B-Base, Cost=15%2026.057.5
Logits Extrap.Backbone=Qwen3-4B-Base, Cost=15%2026.056.7
BaseBackbone=Qwen3-8B-Base, Cost=0%2026.052.9
RELEXBackbone=Qwen2.5-Math-1.5B, Cost=15%2026.052.1
AlphaRLBackbone=Qwen2.5-Math-1.5B, Cost=15%2026.051.3
BaseBackbone=Qwen2.5-Math-1.5B, Cost=0%2026.050.8
BaseBackbone=Qwen3-4B-Base, Cost=0%2026.050.8
RLVRBackbone=Qwen2.5-Math-1.5B, Cost=100%2026.050.4
ExPOBackbone=Qwen2.5-Math-1.5B, Cost=15%2026.050.4
Logits Extrap.Backbone=Qwen2.5-Math-1.5B, Cost=15%2026.050.4
Weight Extrap.Backbone=Qwen2.5-Math-1.5B, Cost=15%2026.050