Skip to main content

Start custom evaluation

POST 

/models/custom_eval/start

Starts a custom evaluation run that compares two models head-to-head using the specified evaluation criteria. A judge model scores each pair of outputs as Win, Draw, or Loss for model_1. Results are aggregated into an improvement percentage: (wins − losses) / total × 100. Custom evaluations are billed per token used by the judge model.

Request

Responses

OK