OpenAI's new state-of-the-art image model ships up to 50% faster generation, in-chat sketching, and a Flare/Sunburst split in the API. Announced September 8, 2026.
2026-09-08 · by Moe Ameen
On September 8, 2026, OpenAI released ChatGPT Images 2.5, calling it its new state-of-the-art image model, and began rolling it out to all ChatGPT, ChatGPT Work, and Codex users across desktop, mobile, and web. The headline is speed and control: OpenAI says generation latency is down by up to 50% compared with the prior Images 2.0, alongside more natural lighting, richer textures, better preservation of subjects from a reference photo, and more reliable editing across multiple turns of a conversation — selective edits that leave the surrounding image intact, and a style that holds steady over a longer chat.
Two features are new inside ChatGPT. Sketch lets you draw directly in the chat, by typing @Sketch, and use that rough drawing as a reference for the final image — closing the gap between describing a composition and roughing it out. Templates give prebuilt starting points for common formats like flyers, posters, and product photos, lowering the floor for people who are not designers.
For developers, OpenAI added two models to the Images API. GPT-Image-2.5 Flare is positioned as the default for most applications — creator and social content, product imagery, visual search, rapid prototyping, and high-volume generation — carrying the quality and editing gains at roughly 50% lower latency. GPT-Image-2.5 Sunburst is built for premium workflows that need tighter control across edits, such as campaign creative and polished product shots, and takes longer to generate. Reported API pricing is token-based, in line with the prior gpt-image generation; exact token usage per image is not documented, so confirm current rates on OpenAI's pricing page before you budget around volume.
There are two plays here, one fast and one durable. The fast one: "OpenAI just made its image model faster and more controllable" is a high-intent search this week, and a single point of view can become a week of content in [Kompozy](/) — a blog explaining what Sketch, Templates, and the Flare/Sunburst split actually change, a [carousel](/glossary/hyperframes) breaking it down slide by slide, short captioned clips, and platform-native posts written in your own voice through your [Persona Brief](/glossary/persona-brief), scheduled across nine platforms plus blog and email. Being early and accurate on a model release is how one opinion becomes a week of output.
The durable one is the fit with what Kompozy already generates. Kompozy's image formats — Photo Posts, Infographic Photo, Carousels, and Quote Graphics — run on OpenAI's gpt-image family, with Google Gemini handling face-locked persona images, so the exact qualities 2.5 sharpens are the qualities those formats trade on. What a chat image model cannot do is the part Kompozy is built for: take one generated image and fan it into a brand-exact carousel via HyperFrames, a quote card, and per-platform posts, keep a persona's face identical across every one with Gemini face-lock, and publish the whole set on a schedule with autopilot and a review step. ChatGPT Images 2.5 makes a better image; Kompozy turns that image into a feed.
It is OpenAI's new image model, announced on September 8, 2026, described as state-of-the-art. It generates and edits images inside ChatGPT with up to 50% lower latency than Images 2.0, more natural lighting, better subject preservation from reference photos, and more reliable editing across a multi-turn conversation. It also adds Sketch and Templates.
It launched on September 8, 2026, rolling out to all ChatGPT, ChatGPT Work, and Codex users across desktop, mobile, and web. Developers reach it through the Images API using two models, GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst.
They are the two API models for Images 2.5. Flare is the default for most uses — creator and social content, product imagery, prototyping, and high-volume generation — at roughly 50% lower latency. Sunburst is built for premium workflows needing tighter control across edits, like campaign creative, and takes longer to generate.
No. It generates and edits images inside a chat, but it does not caption for social, turn one image into a multi-format branded set, or publish anywhere. That distribution step is separate — a tool like Kompozy takes a generated image and fans it into a carousel, quote card, and per-platform posts, then schedules them across nine platforms.