OpenAI ships ChatGPT Images 2.5 (GPT-Image-2.5): @Sketch drawing, Flare and Sunburst API variants, and new pricing tiers

OpenAI released ChatGPT Images 2.5 on 2026-09-08, adding an @Sketch in-chat drawing feature, inline region comments for targeted edits, and two API variants - gpt-image-2.5-flare (fast generation) and gpt-image-2.5-sunburst (precision editing) - with claimed…

On 2026-09-08, OpenAI announced ChatGPT Images 2.5, an update to image generation in ChatGPT that the company says turns ideas, sketches, and reference photos into more personalized, polished images that better reflect user intent; OpenAI's own announcement included no benchmark numbers, model sizes, or architecture details. The Verge reports the release adds a Sketch feature, invoked by typing @Sketch, that opens an in-chat drawing window so user doodles can serve directly as generation prompts, plus inline comments on parts of an image to request targeted edits; The Decoder additionally lists shareable prompts among the new ChatGPT features. OpenAI claims more natural lighting, richer textures, better multi-turn instruction following, and up to 50% latency reduction versus the previous generation, though the two outlets citing the figure name the baseline slightly differently: The Verge says compared with Images 2.0 and The Decoder says versus GPT-Image-2. Availability is listed for ChatGPT, ChatGPT Work, and Codex users on desktop, mobile, and web. Simon Willison's report and The Decoder identify two API model IDs - gpt-image-2.5-sunburst for precision editing and gpt-image-2.5-flare for fast everyday generation - and Willison cites OpenAI's figure of more than 3 billion images generated across ChatGPT Images and the GPT-Image API. The Decoder adds commercial details absent from the other reports: both variants cost $8 per million input tokens and $30 per million output tokens, with new xhigh and max quality tiers (a max-tier 1024x1024 image runs roughly $0.21); it characterizes Flare as the faster default with higher quality than GPT-Image-2 and describes Sunburst as built for precise multi-round edits 'at higher cost', while listing identical per-token pricing for both. In hands-on testing, The Decoder found edit consistency strong in ChatGPT Work but inconsistent in Chat, and notes OpenAI has not documented how ChatGPT routes users between the models. Naming differs across reports: OpenAI and The Verge call the release ChatGPT Images 2.5, while The Decoder refers to the model as GPT-Image-2.5. Beyond these naming and baseline differences, the reports do not contradict each other.

• OpenAI announced ChatGPT Images 2.5 on 2026-09-08; the announcement disclosed no benchmark numbers, model sizes, or architecture details (OpenAI).
• A Sketch feature, invoked by typing @Sketch, opens an in-chat drawing window so user doodles can serve as generation prompts (The Verge).
• Users can leave inline comments on parts of an image to request targeted edits (The Verge); The Decoder also lists shareable prompts among the new ChatGPT features.
• Two API model IDs are exposed: gpt-image-2.5-sunburst for precision editing and gpt-image-2.5-flare for fast everyday generation (Simon Willison; The Decoder).
• OpenAI claims more natural lighting, richer textures, better multi-turn instruction following, and better preservation of subjects from reference photos (OpenAI claims via The Verge and Simon Willison).
• OpenAI claims up to 50% latency reduction versus the prior generation; The Verge cites the baseline as Images 2.0 and The Decoder as GPT-Image-2.
• The Decoder reports both variants cost $8 per million input tokens and $30 per million output tokens, with new xhigh and max quality tiers; a max-tier 1024x1024 image runs roughly $0.21.
• The Decoder characterizes Flare as the faster default with higher quality than GPT-Image-2 and Sunburst as built for precise, multi-round edits 'at higher cost', while listing identical per-token pricing for both variants.