What is the recommended token size (gpt-4o-mini) of the full prompt in the assistant. Under 1000 or under 2500 ? can anyone shed some light on this please ?