OpenAI announced ChatGPT Images 2.5 on September 8, 2026. Alongside the model update, sketching, templates and image comments offer more direct ways to specify changes. The rollout covers ChatGPT, ChatGPT Work and Codex across tiers, while developers can access the Flare and Sunburst API models.
Keeping the rest of an image intact
OpenAI reports stronger reference fidelity and editing consistency, with generation latency reduced by up to 50% against Images 2.0. That is a company claim, not a guarantee for every request or a benchmark conducted for this article.
For a creator, the useful test is the cost of revision. A product-background edit should preserve the label and shape; changing an outfit should preserve the face. These are proposed acceptance tests, not claims that we ran them. Check lettering, identity and brand details before approving an asset.
Sketch first, then describe the details
On mobile, type @ in the message box and choose Sketch. Draw the rough arrangement, then describe the intended content and style. It is another input method when relative positions are hard to explain in prose, not a finished design produced merely by drawing a few lines.
Templates in Images provide starting points such as posters and product photos. On mobile, open a generated image full-screen and add a comment to specify an edit. Shared prompts let another person adapt an idea using their own content rather than simply receiving the finished image.
Availability is not identical across modes
Access across tiers does not mean feature parity. The September 8 release notes explicitly say Templates are not yet available in Work mode and existing image-generation limits remain unchanged. Free accounts are included in the rollout, but unlimited generation was not announced.
If a new control is missing, distinguish rollout timing from mode-specific restrictions. Existing upload-and-edit tools remain an entry point. To judge suitability for posters, comics or product photos, compare the same source material and requested edits instead of treating launch examples as proof.
Flare, Sunburst and API pricing
The API offers GPT-Image-2.5 Flare for most applications and GPT-Image-2.5 Sunburst for more precise creative work. OpenAI positions Flare around quality and speed, while Sunburst takes longer. A practical choice must account for acceptable waiting time as well as output quality.
As checked on September 9, standard API rates are identical: $5 per million text-input tokens, $8 per million image-input tokens and $30 per million image-output tokens. Cached text and image inputs cost $1.25 and $2 per million respectively. These are token rates—not $30 per image. Actual usage determines the bill, so identical rates do not guarantee identical per-image costs. Include revision attempts when estimating a product's costs.
