List your company

Mathematical Reasoning on MATH-500 (Average@16)

88.5Average@16GRPO
47.83658.39368.9579.507Jan 26, 2026

Evaluation results

MethodLinks
GRPOBackbone=Qwen3-4B (non-think)2026.0188.5
Ours (Shorten Prefix)Backbone=Qwen3-4B (non-think)2026.0188
RTSBackbone=Qwen3-4B (non-think)2026.0187.8
MixedBackbone=Qwen3-4B (non-think)2026.0186.2
Ours (Shorten Prefix)Backbone=Qwen3-8B-Base2026.0186
GRPOBackbone=Qwen3-8B-Base2026.0185.2
RTSBackbone=Qwen3-8B-Base2026.0183.4
Ours (Shorten Prefix)Backbone=Qwen3-4B-Base2026.0183
BaseBackbone=Qwen3-4B (non-think)2026.0182.5
GRPOBackbone=Qwen3-4B-Base2026.0182.3
MixedBackbone=Qwen3-8B-Base2026.0182.2
RTSBackbone=Qwen3-4B-Base2026.0180.6
MixedBackbone=Qwen3-4B-Base2026.0180.4
BaseBackbone=Qwen3-4B-Base2026.0158.7
BaseBackbone=Qwen3-8B-Base2026.0149.4