GPT Image 2.5 API: Flare runs 50% faster, Sunburst edits surgically
Both new models drop at GPT Image 2's token rates, add xhigh and max quality tiers, and run through the Images and Responses APIs — the upgrade costs nothing extra.
Copy markdown
Flare: the new default, 50% faster
gpt-image-2.5-flare takes over from GPT Image 2 as the go-to for high-volume work — OpenAI clocks it at 50% lower latency, and Manus's eval team measured 2–4x faster generation. The quality jump (richer textures, natural lighting, better reference-photo fidelity) rides along on both models.
Sunburst edits only what you point at
gpt-image-2.5-sunburst is the precision tier: it changes the region you asked for and leaves the rest of the frame untouched, holding steady across multi-turn edits instead of drifting. That's the difference between a programmatic product-image pipeline you can trust and one that silently rebuilds the whole asset every call.
Two APIs, two new quality tiers
Call either model through the Images API (v1/images) or the Responses API image-generation tool (v1/responses). Both add new xhigh and max quality settings on top of the old options — more headroom when you need print-resolution or campaign-grade output.
The token math
No new pricing tier — both bill at GPT Image 2 rates: $30 per 1M image-output tokens, $8 per 1M image input ($2 cached), $5 per 1M text input. Faster and sharper at the exact budget you already modeled.
Build this weekend
Point Sunburst at a product catalog and script one edit — new background, swapped label, seasonal palette — across every SKU while preserving each shot exactly. Or wire Flare into a live preview loop where users watch variations appear as fast as they type. The precise-edit-plus-speed combo makes both genuinely practical for the first time.
Elsewhere: Sketch and Templates hit ChatGPT
The same launch adds an @Sketch tool (draw a rough layout as a reference), reusable Templates for posters and product shots, and inline image comments — rolling out across ChatGPT, ChatGPT Work, and Codex on every tier.