OpenAI began rolling out ChatGPT Images 2.5 on September 8 to all tiers of ChatGPT, ChatGPT Work and Codex on web, desktop and mobile. According to OpenAI's launch announcement, the updated system cuts generation latency by up to 50 percent compared to Images 2.0.
Sketch an idea, then refine it
Creators can place comments directly on an image to request focused changes. OpenAI says the model is better at changing the requested element while retaining surrounding details and recognizable subjects.
The new Sketch feature, accessed by typing @Sketch, lets users draw a reference for a composition inside ChatGPT. Templates offer starting formats, while shared prompts let others reuse an idea with their own photos. The system card also describes better infographic accuracy and layout.
Flare for speed, Sunburst for precision
Developers can choose two models that accept text and image inputs and produce images:
gpt-image-2.5-flare: OpenAI’s option for fast, everyday image generation.gpt-image-2.5-sunburst: Positioned for more precise editing, with longer generation times.
Both work through the Image API or the Responses API image generation tool, with quality settings from low through max and an auto option. Developer accounts require paid access; the free API tier is unsupported, with Tier 1 accounts subject to a limit of 5 images per minute.
Both models list the same token rates. In OpenAI's official developer pricing table, uncached rates are set at $5 per million text input tokens, $8 per million image input tokens, and $30 per million image output tokens. These are token rates, not fixed prices per image; matching rates do not establish an identical bill for each finished image.
OpenAI’s system card describes C2PA provenance metadata and Google DeepMind’s SynthID invisible watermarking across ChatGPT, Codex and the API.