"GPT 5.4 is currently the leader on our internal benchmarks. Our engineers find it to be more natural and assertive than previous models. It works through ambiguous problems without second-guessing itself and it is proactive in paralyzing work to keep things moving."