List your company
ResearchFormerly WizWand

General Reasoning on MMLU (Accuracy)

96.2MMLU General Reasoning AccuracyCOSE
33.165649.530365.89582.2597May 27, 2026May 28, 2026May 29, 2026May 30, 2026May 31, 2026Jun 1, 2026
Updated 5mo ago

Evaluation results

MethodLinks
COSEModel=Qwen3-4B2026.0596.2
COSEModel=Qwen3-0.6B2026.0595
MAEModel=Llama-3.2-3B-Instruct2026.0595
COSEModel=Llama-3.2-3B-Instruct2026.0594.8
MAEModel=Qwen3-4B2026.0594.2
R-ZeroModel=Qwen3-4B2026.0593.8
AZRModel=Qwen3-4B2026.0593.4
BaseModel=Qwen3-4B2026.0593
COSEModel=Qwen2.5-3B-Instruct2026.0567.5
SUBFITModel=Llama-3.1-8B-Instruct, Sparsity=25%2026.0667.27
SUBFITModel=Qwen3-4B-Instruct, Sparsity=25%2026.0664.23
BaseModel=Qwen2.5-3B-Instruct2026.0563.4
AZRModel=Qwen2.5-3B-Instruct2026.0563.4
R-ZeroModel=Qwen2.5-3B-Instruct2026.0562.8
MAEModel=Qwen2.5-3B-Instruct2026.0561.4
SUBFITModel=Llama-3.2-3B-Instruct, Sparsity=25%2026.0660.69
R-ZeroModel=Llama-3.2-3B-Instruct2026.0558.2
AZRModel=Llama-3.2-3B-Instruct2026.0557.6
MAEModel=Qwen3-0.6B2026.0555.2
BaseModel=Llama-3.2-3B-Instruct2026.0554
R-ZeroModel=Qwen3-0.6B2026.0553.6
AZRModel=Qwen3-0.6B2026.0552.4
SUBFITModel=DeepSeek-7B-chat, Sparsity=25%2026.0649.96
BaseModel=Qwen3-0.6B2026.0544
SUBFITModel=Qwen2.5-7B-Instruct, Sparsity=25%2026.0635.59