Which models are the best evaluators? We're doing internal competition prototypes at @recallnet for subjective skills and finding GPT-4.1 mini to be a surprisingly capable evaluator. The cheap model has taste!