Same here, the same query is significantly slower on chat completion API vs LeChat (roughly 10s vs 1s). I'm using json_object response_format parameter, tried with mistral ai sdk as well as direct curl call. Signaled at Mistral support team today.