List your company

Logical Reasoning on BBH (Accuracy, Time, Peak Memory)

81.62AccuracyVanilla + H2O
27.404841.479955.55569.6301Jun 2, 2026

Evaluation results

MethodLinks
Vanilla + H2OBackbone=Llama3.1-8B Series2026.0681.6223.811,024
VanillaBackbone=Qwen2.5-7B Series2026.0681.0110.592,212
HybridThinkerBackbone=Llama3.1-8B Series2026.068011.731,031
VanillaBackbone=Llama3.1-8B Series2026.0678.1813.812,515
Vanilla + H2OBackbone=Qwen2.5-7B Series2026.0675.7625.011,024
HybridThinkerBackbone=Qwen2.5-7B Series2026.0673.139.871,058
LightThinkerBackbone=Llama3.1-8B Series2026.0672.9313.58999
Vanilla + SepLLMBackbone=Llama3.1-8B Series2026.0672.3215.481,024
CoTBackbone=Qwen2.5-7B Series2026.0671.311.27582
Vanilla + SepLLMBackbone=Qwen2.5-7B Series2026.0669.728.361,024
LightThinkerBackbone=Qwen2.5-7B Series2026.0666.4610.91980
CoTBackbone=Llama3.1-8B Series2026.0653.131.39596
Distill-R1Backbone=Qwen2.5-7B Series2026.0651.313.71905
Distill-R1Backbone=Llama3.1-8B Series2026.0629.495.151,146