List your company

Mathematical Reasoning on GSM8K (Acc, Time, Peak)

91.28AccuracyVanilla + SepLLM
56.273665.361874.4583.5382Jun 2, 2026

Evaluation results

MethodLinks
Vanilla + SepLLMBackbone=Qwen2.5-7B Series2026.0691.2849.031,024
Vanilla + H2OBackbone=Qwen2.5-7B Series2026.0691.0537.981,024
VanillaBackbone=Qwen2.5-7B Series2026.0690.8322.191,559
HybridThinkerBackbone=Qwen2.5-7B Series2026.0690.620.85864
VanillaBackbone=Llama3.1-8B Series2026.0689.3127.891,702
Vanilla + H2OBackbone=Llama3.1-8B Series2026.0689.3148.081,024
CoTBackbone=Qwen2.5-7B Series2026.0688.323.49519
Vanilla + SepLLMBackbone=Llama3.1-8B Series2026.0688.3227.921,024
HybridThinkerBackbone=Llama3.1-8B Series2026.0687.3425.5878
LightThinkerBackbone=Qwen2.5-7B Series2026.0687.2623.05684
LightThinkerBackbone=Llama3.1-8B Series2026.0685.8225.87664
CoTBackbone=Llama3.1-8B Series2026.0664.523.63497
Distill-R1Backbone=Qwen2.5-7B Series2026.0660.586.54522
Distill-R1Backbone=Llama3.1-8B Series2026.0657.627.2524