OpenAI ships GPT Image 2.5: Sunburst for edits, Flare for fast gen
Both cut latency up to 50% at GPT Image 2 prices; plus Claude Code weekly limits drop Sept 14 and OpenAI's prompt-cache diagnostics hit GA.
Copy markdown
Sunburst vs Flare: pick by the job
Sunburst is OpenAI's most capable image model, tuned for precise local edits that keep the approved parts of a frame intact; Flare targets fast drafts and iteration. Both add xhigh and max quality tiers and claim up to 50% lower latency than Images 2.0.
Same token price as GPT Image 2
Pricing is token-based, not per-image: $5/1M text in, $8/1M image in, $30/1M image out (cached $1.25 text / $2 image). OpenAI warns old Image 2 calculators undercount 2.5 usage, so re-estimate before you wire it into production.
Build this weekend: a region-precise photo editor
Endpoints span images/generations, images/edits, Chat Completions and Responses, with new Sketch layout and image-comment tools to direct changes. Enough to ship a product-photo retoucher or campaign-variant generator that only touches the region you point at.
Heads-up: Claude Code weekly limits drop Sept 14
The May '+50%' boost ends; the new permanent level sits at +25% over the old baseline — a 17% cut versus what you have today. It hits Pro, Max, Team and seat-based Enterprise; free and usage-based plans are untouched. Budget your week around it.
OpenAI ships Prompt Cache Diagnostics
Now GA in the Responses API for GPT-5.6 and later, it lets you compare cache performance across responses and debug cache misses — a direct lever on latency and token cost for high-volume apps.