List your company

Language Understanding on MMLU (Accuracy and Refusal Rate)

74.7MMLU AccuracyOriginal
21.86835.58449.363.016May 26, 2026

Evaluation results

MethodLinks
OriginalModel Backbone=Qwen3-14B2026.0574.7—
ICCU (end-to-end)Model Backbone=Qwen3-14B2026.0573.32.8
ICCU (filter + generate)Model Backbone=Qwen3-14B2026.0573.13.3
O3Model Backbone=Qwen3-14B2026.0561.9—
OriginalModel Backbone=Llama-3-8B-Instruct2026.0560.6—
ICCU (filter + generate)Model Backbone=Llama-3-8B-Instruct2026.0560.63.1
RMUModel Backbone=Qwen3-14B2026.0559.9—
ICCU (end-to-end)Model Backbone=Llama-3-8B-Instruct2026.0558.45
O3Model Backbone=Llama-3-8B-Instruct2026.0549.8—
GAModel Backbone=Llama-3-8B-Instruct2026.0530.2—
RMUModel Backbone=Llama-3-8B-Instruct2026.0526.4—
GAModel Backbone=Qwen3-14B2026.0523.9—