List your company

Mathematical Reasoning on AMC 23 (Testing accuracy)

57.5Testing AccuracyGAPG
15.926.737.548.3Sep 28, 2025Nov 12, 2025Dec 27, 2025Feb 10, 2026Mar 27, 2026May 11, 2026Jun 26, 2026

Evaluation results

MethodLinks
GAPGModel=Qwen2.5-1.5B2025.0957.5
Router-R1Model=Qwen2.5-1.5B2025.0952.5
GAPGModel=LLaMA-3.2-3B2025.0952.5
Router-R1Model=LLaMA-3.2-3B2025.0947.5
DynaSteerBackbone=Qwen3-1.7B2026.0647.5
AutoMixModel=Qwen2.5-1.5B2025.0942.5
AutoMixModel=LLaMA-3.2-3B2025.0935
TruthXBackbone=Qwen3-1.7B2026.0635
TruthFlowBackbone=Qwen3-1.7B2026.0635
RLDFGeneration Steps=512, Unmasking Strategy=dynamic, Unmasking Threshold=0.9, Base Model=Dream2026.0530
RLDFGeneration Steps=128, Unmasking Strategy=dynamic, Unmasking Threshold=0.9, Base Model=Dream2026.0517.5