llm-evals ranks models by a weighted score across real intake cases. DeepSeek won overall on Qualified.At, but every dataset crowned a different winner. There is no best model, only the best model for your task.