OpenAI shipped ChatGPT Images 2.5 on September 8, cutting image generation latency by as much as half compared with Images 2.0 and adding a drawing canvas that lets users sketch a reference directly inside ChatGPT. The update reached every ChatGPT tier the same day, five months after Images 2 and a week after the GPT-6 Astra launch.
Key takeaways
- OpenAI claims up to 50% lower generation latency than Images 2.0, alongside more natural lighting, richer textures and better subject preservation from reference photos.
- Three new interface features ship with it: Sketch for drawing a reference in-app, Templates for common formats like flyers and product photos, and comments pinned directly onto an image for targeted edits.
- Developers get two API models β GPT-Image-2.5 Flare as the default and GPT-Image-2.5 Sunburst for precision work that tolerates longer generation times.
What changed in the model
The headline improvements are about control rather than raw capability. According to OpenAI's announcement post, Images 2.5 renders more natural lighting and richer textures, holds onto the subjects in a reference photo more faithfully, and follows editing instructions more reliably when a user keeps refining across multiple turns. That last point matters most in practice: multi-turn drift, where a model quietly loses the face or product it was told to preserve, has been the standing complaint about conversational image editing.
Speed is the other axis. Halving latency changes what the tool is for. Iterating on a concept becomes a conversation rather than a series of waits, which is the premise the new interface features are built on.
The three new tools
Sketch lets users draw directly in ChatGPT and hand that drawing to the model as a compositional reference. It closes a gap that text prompting has never handled well β saying where things go. Templates give a starting point for the formats people generate most, flyers and product photography among them. Comments can now be placed on a specific region of an image, so an edit request attaches to a location instead of being described in prose. Users can also share the prompts they used, letting others rerun the same idea against their own photos and details.
Taken together these are the fixtures of a design tool, not a chat feature, and that is clearly the direction. Images 2 had already added 2K output, multiple aspect ratios, web research for current references, non-Latin language handling and the Instant and Thinking intelligence tiers.
What developers get
The API split is the part worth reading carefully. GPT-Image-2.5 Flare carries the same quality and speed gains and is positioned as the default for most applications, delivering better output than GPT-Image-2 at 50% lower latency. OpenAI points it at creator and social content, product experiences, visual search, rapid prototyping and high-volume generation.
GPT-Image-2.5 Sunburst is the opposite trade. It targets premium visual workflows that need tighter control across successive edits β production campaign creative, polished product imagery β and accepts longer generation times to get there. As 9to5Mac observed, the naming is new but the pattern is familiar: a fast default and a slower, more deliberate sibling.
Outlook
Rolling a multimodal upgrade out to all tiers on day one, including ChatGPT Work and Codex, signals that OpenAI now treats image editing as table stakes rather than a premium differentiator. It is the same cadence the company showed when it shipped two products in a single day earlier this year. The pressure now moves to dedicated design tools, which have to justify a subscription against a generative AI canvas that is bundled, fast and already open on most desks.
FAQ
Who can use ChatGPT Images 2.5?
OpenAI rolled it out to ChatGPT, ChatGPT Work and Codex users across all tiers on desktop, mobile and web on launch day. There is no separate paid upgrade required to reach the new model.
What is the difference between Flare and Sunburst?
Flare is the default API model, matching Images 2.5 quality at roughly half the latency of GPT-Image-2, and suits high-volume or interactive use. Sunburst is built for precision across repeated edits in premium creative workflows and takes longer to generate each image.
What does the Sketch feature do?
Sketch adds a drawing surface inside ChatGPT. A rough drawing becomes a reference the model uses for the final image, which gives users control over layout and composition that text prompts alone handle poorly.
