List your company

Arithmetic Reasoning on GSM8K Aug

83.1AccuracyCoT-SFT
4.0624.5845.165.62Jun 29, 2026

Evaluation results

MethodLinks
CoT-SFTBackbone=Qwen3-4B, Evaluation Protocol=zero-shot2026.0683.1119.8
CoT-SFTBackbone=Qwen3-VL-4B-Instruct, Evaluation Protocol=zero-shot2026.0681.2127.3
DLR (Full)Backbone=Qwen3-VL-4B-Instruct, Evaluation Protocol=zero-shot2026.0663.36.5
DLR (Full)Backbone=Qwen3-4B, Evaluation Protocol=zero-shot2026.0662.26.5
DLR (SFT)Backbone=Qwen3-VL-4B-Instruct, Evaluation Protocol=zero-shot2026.06486.3
DLR (SFT)Backbone=Qwen3-4B, Evaluation Protocol=zero-shot2026.0647.46.4
CoLaRBackbone=Qwen3-4B, Evaluation Protocol=zero-shot2026.064039.6
CoLaRBackbone=Qwen3-VL-4B-Instruct, Evaluation Protocol=zero-shot2026.0639.238.8
RoTBackbone=Qwen3-VL-4B-Instruct, Evaluation Protocol=zero-shot2026.0637.832
Direct-SFTBackbone=Qwen3-4B, Evaluation Protocol=zero-shot2026.0627.40
Direct-SFTBackbone=Qwen3-VL-4B-Instruct, Evaluation Protocol=zero-shot2026.0626.20
CoconutBackbone=Qwen3-4B, Evaluation Protocol=zero-shot2026.0616.96
CoconutBackbone=Qwen3-VL-4B-Instruct, Evaluation Protocol=zero-shot2026.0616.16
iCoTBackbone=Qwen3-4B, Evaluation Protocol=zero-shot2026.0613.50
CODIBackbone=Qwen3-4B, Evaluation Protocol=zero-shot2026.067.36
CODIBackbone=Qwen3-VL-4B-Instruct, Evaluation Protocol=zero-shot2026.067.16