List your company

Arithmetic Reasoning on MultiArith (Accuracy and #L)

98.5AccuracyCoT-SFT
14.57236.36158.1579.939Jun 29, 2026

Evaluation results

MethodLinks
CoT-SFTBackbone=Qwen3-4B2026.0698.557.6
CoT-SFTBackbone=Qwen3-VL-4B-Instruct2026.0698.359.1
DLR (Full)Backbone=Qwen3-VL-4B-Instruct2026.0698.34.3
DLR (Full)Backbone=Qwen3-4B2026.0697.84.3
DLR (SFT)Backbone=Qwen3-VL-4B-Instruct2026.0697.84.2
RoTBackbone=Qwen3-VL-4B-Instruct2026.0697.232
DLR (SFT)Backbone=Qwen3-4B2026.0696.14.2
Direct-SFTBackbone=Qwen3-4B2026.0686.30
Direct-SFTBackbone=Qwen3-VL-4B-Instruct2026.0685.60
CoLaRBackbone=Qwen3-4B2026.0682.221.1
CoLaRBackbone=Qwen3-VL-4B-Instruct2026.068122.1
CoconutBackbone=Qwen3-4B2026.0660.36
CoconutBackbone=Qwen3-VL-4B-Instruct2026.0658.96
iCoTBackbone=Qwen3-4B2026.0649.20
CODIBackbone=Qwen3-4B2026.0618.36
CODIBackbone=Qwen3-VL-4B-Instruct2026.0617.86