OpenAI’s new Flare and Sunburst models prioritize speed and precision, respectively, but their real-world latency and per-job cost differences remain unclear.