OpenAI launches GPT-Image-2.5 (Flare and Sunburst) with faster generation, new quality tiers, revised pricing, and new ChatGPT editing features

OpenAI replaced GPT-Image-2 with two new models, GPT-Image-2.5 Flare and Sunburst, across ChatGPT (Chat, Work, Codex, free tier) and the API. The update adds two new quality tiers ("xhigh" and "max"), revises pricing to roughly $0.006 (low) to $0.21 (max) per 1024x1024 image, removes the batch discount that existed under the previous model, and adds ChatGPT features including a sketch-to-image tool, prompt templates and SynthID invisible watermarking via a Google DeepMind partnership. No primary source links or documentation were captured in this extraction; the claims trace back to OpenAI's own announcement as relayed by The Decoder.

Anyone budgeting for image generation at scale needs to recheck costs: the removal of the batch discount could raise effective prices for high-volume users even as per-tier headline rates look competitive. Developers and enterprises using the API get new quality options but face a forced migration off GPT-Image-2, and the SynthID/C2PA watermarking addition matters for provenance-sensitive use cases (stock content, journalism, ad compliance). Impact on casual ChatGPT users is minor, this is mostly a backend and pricing change dressed up with UI additions.

The quality claims (sharper details, better subject preservation, 50% lower latency, Arena leaderboard win) are OpenAI's own assertions, not independently verified in this extraction, and Arena leaderboard position is a self-selected benchmark OpenAI has incentive to top. The Decoder's own testing reportedly found inconsistent results and could not pin down an average per-image cost or explain undisclosed routing behaviour in ChatGPT, details omitted from the vendor framing. Treat "faster" and "more precise" as marketing language until independently reproduced.

1. 01Independent benchmark comparisons (not Arena) of GPT-Image-2.5 vs GPT-Image-2 and competitors like Midjourney or Google's Imagen, run by a third party
2. 02Actual API billing data from developers over the next 1-2 months showing whether removing the batch discount raises real-world costs for high-volume users
3. 03Whether SynthID watermarking becomes a factor in platform or regulatory compliance decisions (e.g. ad networks, stock media sites) within the next quarter