Hi, by adding provider the way you have mentioned will work only for LLM output generation and not for evaluation. If you run command promptfoo eval --verbose you will see in the terminal promptfoo is using the mentioned providers only for generation. And, it is using by default open ai model for evaluation i.e. two models named gpt:4o and gpt-4o-mini for generation and evaluation will happen using only one model and that would be by default open ai model. Please let me know if this is not clear.