List your company

Multitask Language Understanding on MMLU f_con o=5

99.5Accuracy (Exact)W/O Decontamination
3.71628.58353.4578.317Jan 27, 2026

Evaluation results

MethodLinks
W/O DecontaminationModel=Mistral, Access=–2026.0199.554.497.253.846.35.7
ITDModel=Mistral, Access=Black-box2026.0199.454.396.753.346.76.1
W/O DecontaminationModel=LLaMA-3, Access=–2026.0199.241.696.139.564.19.2
ITDModel=Qwen 2.5, Access=Black-box2026.0198.331.393.826.775.27.9
W/O DecontaminationModel=Qwen 2.5, Access=–2026.0198.131.194.227.1746.7
TEDModel=Qwen 2.5, Access=Black-box2026.01872083.716.688.621.3
ITDModel=LLaMA-3, Access=Black-box2026.0182.424.882.826.3616.1
DeconIEPModel=Qwen 2.5, Access=White-box2026.0180.513.572.55.464.13.2
TEDModel=LLaMA-3, Access=Black-box2026.0179.421.878.522616.1
Short CircuitModel=Qwen 2.5, Access=White-box2026.0175875.38.270.73.4
DeconIEPModel=LLaMA-3, Access=White-box2026.0173.61652.73.852.72.2
Shortcut NeuronModel=LLaMA-3, Access=White-box2026.0170.51368.51242.712.2
Short CircuitModel=LLaMA-3, Access=White-box2026.0169.712.169.51370.615.7
Short CircuitModel=Mistral, Access=White-box2026.0158.11353.510.140.70.1
DeconIEPModel=Mistral, Access=White-box2026.0145.40.342.31.143.32.7
Shortcut NeuronModel=Qwen 2.5, Access=White-box2026.0140.526.541.925.235.931.4
Shortcut NeuronModel=Mistral, Access=White-box2026.0139.45.740.92.524.516.1
TEDModel=Mistral, Access=Black-box2026.017.437.732.41138.32.3