From the source
OpenAI introduced GPT Images 2.5, offering two new models: Flare, positioned as the faster default option with 50% lower latency than GPT-Image-2, and Sunburst, which provides greater precision and control for premium workflows but with longer generation times.
Both models share the same token rates, but OpenAI does not explain how per-image costs will compare or provide a way to estimate token consumption for the new models.