The 5 seconds delay is probably due to reasoning. Maybe try setting it to minimal? If your use case isn’t complex maybe reasoning is overkill and gpt-4.1 would suffice.