List your company

Mathematical Reasoning on MATH500 In-Domain (test)

86.2AccuracyDynaSteer
44.655.466.277Jun 26, 2026

Evaluation results

MethodLinks
DynaSteerBackbone=Qwen3-8B2026.0686.2
ACTBackbone=Qwen3-8B2026.0684.8
DRESSBackbone=Qwen3-8B2026.0684.6
"Wait"Backbone=Qwen3-8B2026.0684.2
ITIBackbone=Qwen3-8B2026.0684.2
PlainBackbone=Qwen3-8B2026.0684
DynaSteerBackbone=Qwen3-1.7B2026.0674.8
DRESSBackbone=Qwen3-1.7B2026.0672.8
ACTBackbone=Qwen3-1.7B2026.0672.6
ITIBackbone=Qwen3-1.7B2026.0671.8
"Wait"Backbone=Qwen3-1.7B2026.0671.2
PlainBackbone=Qwen3-1.7B2026.0669.2
DynaSteerBackbone=Llama3.2-3B-Instruct2026.0652.4
ACTBackbone=Llama3.2-3B-Instruct2026.0648.8
DRESSBackbone=Llama3.2-3B-Instruct2026.0648.6
ITIBackbone=Llama3.2-3B-Instruct2026.0648.4
PlainBackbone=Llama3.2-3B-Instruct2026.0647.8
"Wait"Backbone=Llama3.2-3B-Instruct2026.0646.2