ChatGPT Images 2.5 is OpenAI's image generation model built into ChatGPT, replacing Images 2.0 with latency cut by up to 50%. Every week, over 3 billion images come out of ChatGPT Images and the GPT-Image API, so even small gains ripple widely. On the free plan you can try it with a capped number of images, since OpenAI has not published per-tier quotas. Developers get two API engines on top, GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst.
Comment-based editing is the change you feel first, you pin a note on the area to fix and the model adjusts that spot without disturbing the rest. Circle the mug in your product shot, type make it red, and the background holds still. A more direct route than the masks and layers common in AI image editing tools.
Sketch rounds out the kit, type @Sketch in a conversation, draw your idea (your doodle can be ugly, that's the point) and it becomes the composition plan. Templates for posters or merch and shareable prompts complete the set. Above all, reference photos survive better, lighting and textures carry through style changes, and details stay consistent across a string of edits.
The rivalry is fierce, Google shipped Nano Banana Pro then Nano Banana 2 in quick succession, Midjourney keeps its aesthetic crown and Flux 2 from Black Forest Labs plays the open card. Published head-to-heads often score Google higher on raw photorealism and OpenAI higher on edit instructions, each has its turf.
OpenAI's card remains integration, everything happens inside a conversation, no sliders and no separate app. Adobe has added the model to Firefly. Every file carries an invisible watermark plus metadata following the C2PA provenance standard.
Both API models launched at the same rate, $8 per million image input tokens, $30 per million on output and $5 per million text tokens. Treat those numbers as a launch snapshot, the live figures sit on OpenAI's pricing page. Inside ChatGPT, the rollout reaches every plan plus ChatGPT Work and Codex, on web, mobile and desktop.
| Access | Positioning | Best for |
|---|---|---|
| ChatGPT (all plans) | Consumer face of Images 2.5 | Everyday images, guided edits |
| GPT-Image-2.5 Flare | Fast, the API default | High volume, prototyping, social content |
| GPT-Image-2.5 Sunburst | Precision, longer generation times | Campaign assets, demanding edit work |
Speed and control lead the list. Latency drops by up to 50% versus Images 2.0, reference photos keep their lighting and textures, edits stay consistent across turns, and ChatGPT gains comment-based editing, the Sketch drawing tool, format templates and shareable prompts.
Yes, the free ChatGPT plan includes it with a capped number of images, and OpenAI has not published per-plan quotas. Paid tiers raise the ceiling, and the model also reaches ChatGPT Work and Codex accounts across web, mobile and desktop.
It depends on the job. Tech press comparisons often favor Google on pure photorealism, while OpenAI leads on edit instructions and keeping a character consistent across a series. If ChatGPT already sits open in your browser all day, Images 2.5 removes most reasons to switch.
Commercial use is allowed, OpenAI's terms assign you rights to your generations as long as usage policies are respected. Bear in mind each file carries C2PA metadata and an invisible watermark identifying it as AI-made, worth knowing before an ad campaign goes out.
Verdict: Half the wait and edits that hold steady across a series, that's the pitch in one line. Marketing teams and designers who rework photos of real people and products will get the most out of this update.
