List your company

Mathematical Reasoning on GSM-Hard (Accuracy, Reasoning Length)

53.4AccuracyCoT-SFT
-0.05613.82227.741.578Jun 15, 2026Jun 17, 2026Jun 19, 2026Jun 22, 2026Jun 24, 2026Jun 26, 2026Jun 29, 2026

Evaluation results

MethodLinks
CoT-SFTBackbone=Qwen3-VL-4B-Instruct2026.0653.4191.1
CoT-SFTBackbone=Qwen3-4B2026.0652.3189.4
CoT-SFTBackbone=Qwen3-VL-2B-Instruct2026.0633.1207.2
DLR (Full)Backbone=Qwen3-4B2026.0619.37
DLR (Full)Backbone=Qwen3-VL-4B-Instruct2026.0618.37
CoT-SFTBackbone=LLaMA-3.2-1B2026.0615.6197.5
DLR (Full)Backbone=Qwen3-VL-2B-Instruct2026.0615.46.3
DLR (SFT)Backbone=Qwen3-VL-4B-Instruct2026.0614.76.8
RoTBackbone=Qwen3-VL-4B-Instruct2026.0614.132
DLR (SFT)Backbone=Qwen3-4B2026.0613.76.8
LTFBackbone=LLaMA-3.1 Instruct 8B2026.0613.193.7
ReGuLaRBackbone=LLaMA-3.1 Instruct 8B2026.0612.533.92
LTFBackbone=LLaMA-3.2 Instruct 3B2026.0611.883.64
ReGuLaRBackbone=LLaMA-3.2 Instruct 3B2026.0611.23.73
DLR (Full)Backbone=LLaMA-3.2-1B2026.06116.6
CoLaRBackbone=LLaMA-3.1 Instruct 8B2026.0610.976.53
CoLaRBackbone=LLaMA-3.2 Instruct 3B2026.069.736.46
Direct-SFTBackbone=Qwen3-VL-4B-Instruct2026.069.50
Direct-SFTBackbone=Qwen3-4B2026.069.20
CoLaRBackbone=Qwen3-4B2026.069.247.4
LTFBackbone=DeepSeek-R1 Distill-Qwen 1.5B2026.068.833.94
LTFBackbone=LLaMA-3.2 Instruct 1B2026.068.713.97
RoTBackbone=Qwen3-VL-2B-Instruct2026.068.632
ReGuLaRBackbone=DeepSeek-R1 Distill-Qwen 1.5B2026.068.334.1
ReGuLaRBackbone=LLaMA-3.2-1B2026.068.34.1
CoconutBackbone=LLaMA-3.1 Instruct 8B2026.068.266
ReGuLaRBackbone=LLaMA-3.2 Instruct 1B2026.068.254.1
CoLaRBackbone=Qwen3-VL-2B-Instruct2026.068.26.5
CoLaRBackbone=Qwen3-VL-4B-Instruct2026.068.247.8
CoconutBackbone=LLaMA-3.2 Instruct 3B2026.067.346
CoLaRBackbone=DeepSeek-R1 Distill-Qwen 1.5B2026.066.327.03
CoLaRBackbone=LLaMA-3.2 Instruct 1B2026.066.227.03
CoLaRBackbone=LLaMA-3.2-1B2026.066.27
SoftCoT++Backbone=LLaMA-3.1 Instruct 8B2026.065.91124.82
iCoTBackbone=LLaMA-3.1 Instruct 8B2026.065.820
CoconutBackbone=Qwen3-4B2026.065.46
SoftCoT++Backbone=LLaMA-3.2 Instruct 3B2026.065.21124.72
CoconutBackbone=Qwen3-VL-4B-Instruct2026.065.26
iCoTBackbone=LLaMA-3.2 Instruct 3B2026.065.190
CoconutBackbone=DeepSeek-R1 Distill-Qwen 1.5B2026.064.966
CoconutBackbone=LLaMA-3.2-1B2026.064.96
CoconutBackbone=LLaMA-3.2 Instruct 1B2026.064.876
Assist-CoTBackbone=LLaMA-3.1 Instruct 8B2026.064.75125.59
Direct-SFTBackbone=Qwen3-VL-2B-Instruct2026.064.70
CoTBackbone=LLaMA-3.1 Instruct 8B2026.064.51121.66
CoconutBackbone=Qwen3-VL-2B-Instruct2026.064.26
Assist-CoTBackbone=LLaMA-3.2 Instruct 3B2026.064.16125.03
Direct-SFTBackbone=LLaMA-3.2-1B2026.064.10
iCoTBackbone=Qwen3-4B2026.064.10
iCoTBackbone=LLaMA-3.2-1B2026.063.90
SoftCoT++Backbone=DeepSeek-R1 Distill-Qwen 1.5B2026.063.86123.72
CoTBackbone=LLaMA-3.2 Instruct 3B2026.063.84121.68
iCoTBackbone=DeepSeek-R1 Distill-Qwen 1.5B2026.063.830
iCoTBackbone=LLaMA-3.2 Instruct 1B2026.063.820
SoftCoT++Backbone=LLaMA-3.2 Instruct 1B2026.063.77123.56
Assist-CoTBackbone=DeepSeek-R1 Distill-Qwen 1.5B2026.063.07126.03
Assist-CoTBackbone=LLaMA-3.2 Instruct 1B2026.063.02125.19
CODIBackbone=LLaMA-3.2-1B2026.0636
CODIBackbone=LLaMA-3.2 Instruct 1B2026.062.946
CoTBackbone=DeepSeek-R1 Distill-Qwen 1.5B2026.062.81120.78
CoTBackbone=LLaMA-3.2 Instruct 1B2026.062.74120.46
CODIBackbone=Qwen3-VL-2B-Instruct2026.062.46
CODIBackbone=Qwen3-4B2026.062.26
CODIBackbone=Qwen3-VL-4B-Instruct2026.0626