OpenAI has introduced Images 2.5, its new state-of-the-art image generation model, extending the capabilities of ChatGPT Images and the GPT-Image family available through the API. The launch arrives as image generation usage continues to scale: users now create more than 3 billion images every week across ChatGPT and API-based workflows. At that volume, the release functions less like a novelty demo and more like an infrastructure update—one intended to make generated imagery dependable enough to sit inside everyday creative and production pipelines.

According to the company, the updated model produces more natural lighting and richer textures, renders real-world information more accurately, and handles complex layouts—including transparent backgrounds—with greater reliability. A central improvement is subject preservation: when working from reference photos, Images 2.5 keeps people, objects, and distinctive features recognizable as they are placed into new settings, styles, and compositions. That capability speaks to one of the persistent friction points in generative imagery, where successive edits can cause faces, products, or brand elements to drift from their source appearance. This fidelity also carries over to API-based workflows, where teams often need variations that remain anchored to an original source asset.

Editing capabilities have been refined as well. The model modifies only the requested element—such as a product, background, or piece of text—while preserving the surrounding composition, subject, and brand treatment. In multi-turn conversations, earlier edits are more likely to remain intact, with each new instruction building on prior changes rather than degrading overall quality. For production environments, this means targeted updates no longer require regenerating entire assets.

Performance has improved significantly: OpenAI reports that generation latency has been reduced by up to 50% compared with Images 2.0. The gain matters for iteration speed, since creative teams typically compare several concepts before settling on a direction.

Images 2.5 is here. I don't think it can solve super difficult math problems, but it is really good and we hope you enjoy it.

— Sam Altman (@sama) September 8, 2026

New Creative Tools and Developer-Focused API Models

Alongside the model upgrade, OpenAI is introducing several new product features in ChatGPT, nudging the assistant toward something closer to a lightweight design tool. Sketch allows users to draw directly in the chat interface and use the drawing as a visual reference for the final image, paired with a text description of the desired style and details. Templates provide structured starting points for popular formats such as posters, flyers, merch, and product photos. Users can also attach comments directly to images for more precise editing instructions, and when sharing an image, they can include the prompt that produced it so others can adapt the idea with their own photos and inputs.

For developers, the API gains two new models, in a pairing that has become a familiar pattern across AI model releases: a fast default alongside a slower, higher-fidelity option. GPT-Image-2.5 Flare serves as the default option for most applications, offering higher quality than GPT-Image-2 at half the latency; it is positioned for creator and social content, product experiences, visual search, and high-volume generation. GPT-Image-2.5 Sunburst targets premium workflows—such as production-ready campaign creative and polished product imagery—where tighter control across edits justifies longer generation times.

Early enterprise adopters, including Higgsfield AI, Adobe, Manus, and Runway, have highlighted the model's ability to make meaningful edits without compromising the character, composition, or visual identity of the original image—a property considered critical for film, user-generated content, and advertising pipelines. Adobe and Runway, in particular, are long-established names in creative software and video generation, and their early presence points to the same film, UGC, and advertising pipelines OpenAI identifies as dependent on edit fidelity.

On the safety side, Images 2.5 retains OpenAI's existing safeguards, including prompt and output checks, C2PA metadata—an open industry standard for recording a piece of media's origin and edit history—and invisible watermarking to identify AI-generated content. Provenance features of this sort carry weight as generated imagery moves through social platforms and commercial supply chains. The model is rolling out to ChatGPT, ChatGPT Work, and Codex users across all tiers on desktop, mobile, and web; with both API variants and all consumer tiers covered, how teams divide work between Flare's speed and Sunburst's control in live production is the detail most worth tracking as adoption widens.