What Is GPT Image 2.5? The Complete Guide (2026)
What is GPT Image 2.5? A clear guide to OpenAI's September 2026 image model family: Flare vs Sunburst, what changed from GPT Image 2, and how to try it.
The short answer: GPT Image 2.5 is OpenAI's current generation of image generation and editing models, announced on September 8, 2026 as the successor to GPT Image 2. In the ChatGPT app it ships as ChatGPT Images 2.5. Through the API it comes in two variants: GPT-Image-2.5 Flare, the fast default for most work, and GPT-Image-2.5 Sunburst, a slower variant built for precision. The headline change is simple to state: OpenAI says the new family produces higher-quality images at up to 50% lower latency than the previous generation.
That combination, better output in less time, is why GPT Image 2.5 matters if you generate or edit images as part of your work. But the model family introduces a few new names and a few real capability changes, and most coverage written before the announcement is a mix of confirmed facts and rumors. This guide sticks to what OpenAI has published, plus practical notes on using the model today.
GPT Image 2.5 at a glance
| Question | Answer |
|---|---|
| What is it? | OpenAI's image generation and editing model family, announced September 8, 2026 |
| In-app name | ChatGPT Images 2.5 (ChatGPT, Work, and Codex, rolling out across all tiers) |
| API variants | GPT-Image-2.5 Flare (fast, default) and GPT-Image-2.5 Sunburst (precision, slower) |
| Replaces | GPT-Image-2 on the API side; Images 2.0 inside ChatGPT |
| Headline claim | Higher-quality images at up to 50% lower latency than Images 2.0 |
| Best at | Reference-based edits, precise local changes, multi-turn editing, complex compositions |
| Access | ChatGPT, the OpenAI API, or third-party studios like this one |
If you only needed the summary, you can stop here. The rest of this guide explains each name, what the model actually does well, how it relates to GPT Image 2, and how to start using it.
The three names, explained
GPT Image 2.5 is a family, not a single model. The naming is easier to follow once you separate where each model lives.
ChatGPT Images 2.5 is the version most people will touch first. It powers image generation inside ChatGPT and is rolling out across all tiers on desktop, mobile, and web, including ChatGPT Work and Codex.
GPT-Image-2.5 Flare is the API model OpenAI describes as "the default choice for most applications." It is the fast option: OpenAI's own announcement says Flare delivers higher-quality images than GPT-Image-2 at 50% lower latency, and an evaluation by Manus measured Flare at two to four times the speed of GPT-Image-2.
GPT-Image-2.5 Sunburst is the second API variant, positioned for "an extra level of precision" on detailed creative work, at the cost of longer generation times.
Choosing between Flare and Sunburst is a workflow decision more than a quality ranking. We cover it in detail in GPT Image 2.5 Flare vs Sunburst: which to use. The short version: Flare for exploration and everyday generation, Sunburst for detail-sensitive final passes.
What GPT Image 2.5 can do
OpenAI's announcement makes specific claims about the family. Here is what they amount to in practice.
Faster generation without a quality trade-off. Latency is down "by up to 50% compared with Images 2.0." Speed changes how you work: exploring five visual directions stops being a waiting game, and iterative editing becomes practical inside a single session.
Better reference-photo fidelity. When you supply a reference image, subjects stay more recognizable, with more natural lighting and richer textures. This is the capability behind the model's strongest editing use cases: product shots that must keep looking like the actual product, and portraits that must keep looking like the actual person.
Precision editing. You can request a change to one element, such as a background, a color, or one object, and the surrounding detail is preserved instead of re-rolled. Earlier image models tended to regenerate the whole image and let unrelated details drift. GPT Image 2.5 is built to change only what you asked for.
Multi-turn consistency. Edits persist across successive turns without quality degrading. In practice, this means you can refine an image over several instructions (adjust the light, then the background, then a small object) and the earlier changes survive each step.
Complex instructions and layouts. The family handles multi-part prompts, accurate real-world content, and complex compositions better than the previous generation, and it can generate images with transparent backgrounds, which is useful for logos, stickers, and layered design work.
New in-app tools. Inside ChatGPT, Images 2.5 adds Sketch (draw a rough visual reference by typing @Sketch), templates for common formats like posters and merch, comments placed directly on images, and prompt sharing alongside shared images.
One honest caveat: none of this removes the final inspection step. Small text, hands, faces, and object counts still deserve a check at the intended display size before you publish anything commercially.
GPT Image 2.5 vs GPT Image 2
If you are coming from the previous generation, the practical question is whether the upgrade is worth it. The short version: OpenAI's positioning is that Flare beats GPT-Image-2 on quality at half the latency, and the editing and consistency improvements are the bigger everyday difference: precision edits and multi-turn sessions were where GPT-Image-2 most often forced rework.
We put the full comparison, including a same-prompt test run through both models, in GPT Image 2.5 vs GPT Image 2: is the upgrade worth it?.
From rumor to release
GPT Image 2.5 arrived with an unusual amount of pre-announcement noise. In the weeks before September 8, coverage consisted mostly of rumor roundups sorting leaks from speculation, and the model circulated under the rumored codename "Luna Lisa" in some reports. That era is over: the model line is official, documented, and available. If you read an older article hedging about whether the model exists, treat it as a snapshot of the rumor phase, not the current state.
How to use GPT Image 2.5
There are three main routes, and they suit different people.
In ChatGPT. The simplest path. Images 2.5 is rolling out to all tiers across desktop, mobile, and web. If you already use ChatGPT, image generation and the new editing tools are built in.
Through the API. Developers can call GPT-Image-2.5 Flare and Sunburst directly. Flare is documented as the default for most applications. Current pricing lives on OpenAI's developer pricing page. Per-image cost varies with quality setting and size, so check the numbers for your actual usage rather than relying on summaries.
In a studio like this one. If you want to generate and edit without setting up an API client, this studio runs GPT Image 2.5 in the browser with starter credits. Pick a variant, write a prompt, upload a reference if you need one, and iterate. For prompt ideas, the GPT Image 2.5 prompts gallery collects tested starting points by category.
What it is good at, and where to stay careful
Strengths, in rough order of how much they change daily work:
- Iterative editing. Multi-turn sessions with persistent edits are the family's signature strength.
- Reference-led work. Product mockups, portraits, and brand-consistent assets benefit from the fidelity improvements.
- Text-in-image drafts. GPT Image 2.5 can render legible text, which makes poster and packaging drafts faster, but treat exact wording as a draft until inspected.
- Speed-sensitive workflows. Social content, concept exploration, and high-volume variations.
Stay careful with: final typesetting (move exact copy into a design tool), fine print at small sizes, and any output used commercially without review. And when a task truly needs a precision pass, that is what Sunburst is for.
FAQ
Is GPT Image 2.5 free?
In ChatGPT, image generation is included in the rolling release across tiers, subject to the plan's usage limits. Via the API you pay per image, with cost depending on quality setting and size. Third-party studios set their own terms; this one offers starter credits so you can try before paying.
When was GPT Image 2.5 released?
OpenAI announced it on September 8, 2026, with a rollout across ChatGPT, ChatGPT Work, and Codex on desktop, mobile, and web.
Is GPT Image 2.5 the same as "Luna Lisa"?
"Luna Lisa" was a rumored codename from pre-announcement leaks. The official model family is GPT Image 2.5, with Flare and Sunburst as the API variants.
What is the difference between Flare and Sunburst?
Flare is the fast default for most generation. Sunburst trades longer generation times for extra precision on detailed work. See the full Flare vs Sunburst comparison.
Is GPT Image 2.5 better than GPT Image 2?
On OpenAI's stated numbers, yes: higher-quality output at up to 50% lower latency, plus stronger editing and consistency. Whether it changes your results enough to migrate depends on your workflow. See the dedicated comparison.
Can GPT Image 2.5 render text in images?
It renders legible text, which is enough for drafts and layouts. For production artwork, keep exact wording short, put it in quotation marks in the prompt, and inspect the result at final size.
Bottom line
GPT Image 2.5 is OpenAI's current image family: faster than GPT-Image-2, noticeably better at reference fidelity, precision edits, and multi-turn consistency, and split into Flare (fast default) and Sunburst (precision) on the API side. If you generate images regularly, the editing workflow alone is worth a test.
Open the studio and try GPT Image 2.5 now. Start with Flare, bring a reference image if you have one, and see how the multi-turn editing holds up on your own brief.