OpenAI confirmed new models cut task completion delays by up to 3.6 times while using rated chip power. Long response times make automated systems and real-time agent loops impractical.