Product Launch
GPT Image 2.5 Release Date and Current Status: Released September 8, 2026
EvoLink Team
Product Team
September 4, 2026
Updated on September 9, 2026
15 min read
GPT Image 2.5 was released on September 8, 2026. OpenAI shipped it as ChatGPT Images 2.5 in the app and as two API models,
gpt-image-2.5-sunburst and gpt-image-2.5-flare, on the same day. The model IDs, quality tiers, rate limits, and token prices are published. EvoLink opened both routes one day later, on September 9, 2026: GPT Image 2.5 Flare and GPT Image 2.5 Sunburst, each with its own playground, price module, and API reference. This page records the release itself and helps you decide what to do next.GPT Image 2.5 status at a glance
| Question | Status on September 9, 2026 | Evidence |
|---|---|---|
| Is GPT Image 2.5 released? | Yes | OpenAI's API changelog entry of September 8, 2026: "Released gpt-image-2.5-sunburst and gpt-image-2.5-flare for image generation and editing" |
| What is the release date? | September 8, 2026, ChatGPT and API together | Changelog, model catalog, pricing page, and the ChatGPT Images 2.5 launch post were all updated that day |
| What is it called? | ChatGPT Images 2.5 (product); GPT-Image-2.5 Sunburst and GPT-Image-2.5 Flare (API models) | OpenAI model catalog |
| Is there a model ID and price? | Yes, at OpenAI | Two model IDs with dated snapshots; token rates identical to gpt-image-2 |
Was mona-lisa-1 GPT Image 2.5? | Still unconfirmed | OpenAI's release materials do not mention the Arena labels; see the Mona Lisa 1 release status and Arena test |
| Is GPT Image 2 deprecated? | No | gpt-image-2 remains listed and priced with no deprecation date |
| Can I call GPT Image 2.5 through EvoLink? | Yes, both routes are live | GPT Image 2.5 Flare and GPT Image 2.5 Sunburst went live on EvoLink on September 9, 2026 |
What OpenAI released on September 8, 2026
OpenAI describes ChatGPT Images 2.5 as its new state-of-the-art image model, rolling out to ChatGPT and ChatGPT Work users across all tiers on desktop, mobile, and web. The API gets two models rather than one:
| API model | OpenAI's description | OpenAI's recommendation |
|---|---|---|
gpt-image-2.5-flare | "Fast, high-quality everyday image generation"; brings the same quality, editing, and speed improvements to the API | The default choice for most applications: creator and social content, product experiences, visual search, rapid prototyping, high-volume generation |
gpt-image-2.5-sunburst | "Our most capable model for image generation and editing", built for workflows where editing precision matters most | Premium visual workflows that benefit from tighter control across edits, such as production-ready campaign creative and polished product imagery, at longer generation times |
Both models take text and image inputs, output images, support generation and editing (including inpainting) through the Images API, and can also be selected as the model of the Responses API image generation tool. Each has an undated ID and a dated snapshot ending in
2026-09-08.ChatGPT also gained Sketch, which lets users draw directly inside ChatGPT as a visual reference for the final image. Sketch is an app feature, not an API parameter.
What changed from GPT Image 2, and what did not
What OpenAI says changed:
- Quality tiers. Both 2.5 models add
xhighandmaxabovehigh, and default toautoon OpenAI. EvoLink supports five explicit tiers and defaults tomedium. Per-image costs depend on model, size, quality, inputs, and actual token usage; equal token rates do not guarantee equal total costs. - Latency. OpenAI reports image generation latency reduced by up to 50% compared with Images 2.0, with Flare as the fast model and Sunburst trading speed for precision. The image generation guide still notes that complex prompts can take up to about two minutes.
- Editing. OpenAI says the release follows editing instructions more reliably across multiple turns and is better at changing only the elements a user asks to edit while leaving the rest intact.
- Fidelity. OpenAI says outputs have more natural lighting and richer textures, and better preserve the subjects in reference photos.
What did not change, from OpenAI's documentation:
- Token rates. Both 2.5 models list the same image-input, cached-input, image-output, and text-input rates as
gpt-image-2. Reports of a 2x price increase compare againstgpt-image-2's batch rates, which have no 2.5 equivalent yet. - Size rules. Custom
WIDTHxHEIGHTvalues must be multiples of 16, with ratios from 1:3 to 3:1, at most 3840 px per edge, and 655,360–8,294,400 total pixels. Resolutions above 2560×1440 are experimental. Transparent output requires PNG or WebP. - Rate limits. The same tiered table as
gpt-image-2, from 5 to 250 images per minute; the free tier is not supported. - Documented limitations. Text rendering, consistency across generations, and precise composition remain listed limitations in the guide.
- GPT Image 2's status.
gpt-image-2is still listed as current with its April snapshot and no deprecation date.
Every claim in the first list is OpenAI's own statement. For a production decision, measure latency, first-pass rate, and cost per accepted image on your verified route against a frozen GPT Image 2 baseline.
Timeline: from GPT Image 1.5 to the release
| Date | Event | What it supports |
|---|---|---|
| December 16, 2025 | OpenAI releases GPT Image 1.5 in ChatGPT and the API | The .5 naming precedent that 2.5 followed |
| April 21, 2026 | OpenAI releases GPT Image 2, snapshot gpt-image-2-2026-04-21 | The production baseline that 2.5 is measured against |
| August 9, 2026 | Community first reports the anonymous label mona-lisa-1 in Arena Battle Mode | An unreleased image checkpoint was being tested |
| August 11, 2026 | EvoLink generates an output from mona-lisa-1 in Arena Battle Mode | A callable anonymous checkpoint existed in that session |
| August 20, 2026 | Community posts describe luna-lisa-alpha; OpenAI adds transparent-background preview to gpt-image-2 | A second test label; OpenAI still shipping features on GPT Image 2 |
| September 3, 2026 | Community members report a feature-gated imagegen25 announcement modal in OpenAI's web client code | Announcement copy for an image upgrade was staged |
| September 4, 2026 | OpenAI's catalog, pricing, changelog, and SDK types still showed no image model newer than gpt-image-2 | The release had not happened at that check |
| September 8, 2026 | OpenAI releases ChatGPT Images 2.5 and the API models gpt-image-2.5-sunburst and gpt-image-2.5-flare | The release, its official names, IDs, and prices |
The cadence ended up as roughly four and a half months from GPT Image 2 to 2.5, close to the four months between 1.5 and 2.
What the pre-release signals turned out to be
The
imagegen25 string reported on September 3, 2026, with its copy "Image creation got a major upgrade" and "Higher-quality results, faster generation, and smarter creative tools", preceded the ChatGPT Images 2.5 release by five days. OpenAI has not commented on the string itself, and matching language is not independent confirmation of its intended use. It remains a historical community report.mona-lisa-1 and luna-lisa-alpha are a different case. OpenAI's launch post, both model pages, and the ChatGPT Images 2.5 system card do not mention either label, and Arena has not disclosed their provider. Whether either was a 2.5 checkpoint remains unconfirmed. The first-hand Arena record, including what EvoLink observed and what it does not prove, stays on the Mona Lisa 1 release status page.
EvoLink’s August 11 Arena session records an anonymous label and one output. It does not establish a connection to Flare or Sunburst, and is not a GPT Image 2.5 benchmark.
Is it GPT Image 2.5, ChatGPT Images 2.5, or Sunburst and Flare?
| What people search | What OpenAI calls it | Status |
|---|---|---|
| GPT Image 2.5, gpt-image-2.5 | GPT Image 2.5 family; API models gpt-image-2.5-sunburst and gpt-image-2.5-flare | Released September 8, 2026 |
| ChatGPT Images 2.5, ChatGPT Image 2.5 | ChatGPT Images 2.5, the product name inside ChatGPT | Released September 8, 2026 |
| GPT Image 3 | Not used by OpenAI for this release | The community's alternative guess did not materialize |
| ChatGPT Image 2, ChatGPT Images 2.0 | gpt-image-2 | Released April 21, 2026; still current, no deprecation date |
| GPT Image 2.5 generator sites | Not OpenAI products | Verify which model ID a third-party site actually calls |
The community guessed the
.5 name correctly from the GPT Image 1.5 precedent. The official release established two API variants, Sunburst and Flare. EvoLink keeps "GPT Image 2.5" as the family name on this page and the status page because it is the term people search, and lists the two official IDs wherever a request string matters.Release, API, and EvoLink route: three dates
Three channels moved on consecutive days. OpenAI released the ChatGPT product and the API on September 8, 2026; EvoLink opened both routes on September 9.
| Channel | What "released" means there | Status on September 9, 2026 |
|---|---|---|
| ChatGPT product | The new image experience appears in the app | Released, rolling out across tiers |
| OpenAI API | Model IDs, prices, and documentation appear in the catalog | Released, two model IDs |
| EvoLink route | Gateway model ID, price in the price module, and production request behavior | Live: gpt-image-2.5-flare on /gpt-image-2-5, gpt-image-2.5-sunburst on /gpt-image-2-5-sunburst |
GPT Image 1.5 and GPT Image 2 also shipped to ChatGPT and the API together, and 2.5 followed the pattern. EvoLink added both routes one day after OpenAI's release. There is no
gpt-image-2.5 alias, so requests must name Flare or Sunburst explicitly; each product page carries its own playground, price module, and API reference.Should you migrate now?
| Your situation | Recommendation |
|---|---|
| You call OpenAI directly and want the upgrade today | Start with gpt-image-2.5-flare at auto quality, since OpenAI positions it as the default; move editing-heavy jobs to gpt-image-2.5-sunburst only where multi-turn precision pays for the longer generation time |
| You run on EvoLink | Both routes are live. Start with GPT Image 2.5 Flare at an explicit quality such as the medium default for everyday generation and test Sunburst on editing-heavy jobs; keep GPT Image 2 as the frozen baseline while you compare |
| You are planning an evaluation | Freeze 30 to 100 representative prompts, source images, and acceptance rules with GPT Image 2 outputs now, then score Sunburst and Flare separately on first-pass rate, cost per accepted image, and p50/p95 latency |
You are tempted by xhigh or max | Measure cost per accepted image with input costs, output usage, and retries included; compare explicit quality settings before adopting a higher tier. |
| You only saw launch samples | Treat them as test ideas; OpenAI's quality and latency figures are the vendor's claims until reproduced on your prompts |
Signals that will change this page
- EvoLink publishing measured latency, usage, and cost per accepted image for Flare and Sunburst against GPT Image 2.
- OpenAI publishing a deprecation date for
gpt-image-2, changes to model-specific cost estimates, batch pricing for the 2.5 models, or a new snapshot. - Arena or OpenAI disclosing the provider or identity of
mona-lisa-1orluna-lisa-alpha. - Reproducible workload evaluations of Sunburst and Flare against GPT Image 2 on verified routes.
EvoLink rechecks OpenAI's model catalog, changelog, pricing page, and image generation guide on every update to this page. The last check date is in the update log below.
Where to go from here
| Your goal | Start here |
|---|---|
| Call Flare through EvoLink: playground, price, API reference | GPT Image 2.5 Flare API |
| Call Sunburst through EvoLink | GPT Image 2.5 Sunburst API |
| Read the first-hand Arena test of the anonymous checkpoint | Mona Lisa 1 release status |
| Keep the previous OpenAI image route as a baseline | GPT Image 2 on EvoLink |
| Compare GPT Image 2.5, 2, and 1.5 | GPT Image family |
| Integration guide with code examples | GPT Image 2 developer guide |
| Decide between GPT Image 2 and 1.5 | GPT Image 2 vs GPT Image 1.5 |
| Browse prompt examples | GPT Image 2 prompts |
Understand the "ChatGPT Image 2" vs gpt-image-2 naming | ChatGPT Image 2 release status |
For a task-based decision between the two variants, use the Flare vs Sunburst guide, including paired test briefs and cost per accepted image.
FAQ
Is GPT Image 2.5 released?
Yes. OpenAI released it on September 8, 2026, as ChatGPT Images 2.5 in the app and as the API models
gpt-image-2.5-sunburst and gpt-image-2.5-flare.What is the GPT Image 2.5 release date?
September 8, 2026, for both ChatGPT and the OpenAI API. The dated API snapshots end in
2026-09-08.What is the difference between Sunburst and Flare?
OpenAI positions Flare as the default for most applications, with higher quality than GPT Image 2 at lower latency, and Sunburst as the precision model for editing-heavy work at longer generation times. They share endpoints, quality tiers, size rules, and list prices.
Is ChatGPT Images 2.5 the same as GPT Image 2.5?
Yes. ChatGPT Images 2.5 is the product name inside ChatGPT; GPT Image 2.5 is the API family, split into the Sunburst and Flare models.
Is GPT Image 2.5 twice the price of GPT Image 2?
No. OpenAI's standard token rates are identical for
gpt-image-2 and both 2.5 models. Comparisons that show a 2x jump are measuring against gpt-image-2's batch rates, which have no 2.5 equivalent yet. Per-image cost still depends on size and quality tier; the GPT Image 2.5 API status page carries the list-price table.What is imagegen25?
A community-reported internal identifier from a feature-gated announcement modal found in OpenAI's web client code on September 3, 2026. The ChatGPT Images 2.5 release followed five days later; OpenAI has not commented on the string itself.
Was Mona Lisa 1 GPT Image 2.5?
Unconfirmed.
mona-lisa-1 is an anonymous Arena label that EvoLink observed directly on August 11, 2026. OpenAI's release materials do not mention it, and Arena has not disclosed its provider.Is GPT Image 2 deprecated now?
No. As of September 8, 2026,
gpt-image-2 remains listed and priced by OpenAI with no deprecation date.Can I use GPT Image 2.5 on EvoLink?
Yes. Both routes went live on EvoLink on September 9, 2026: GPT Image 2.5 Flare and GPT Image 2.5 Sunburst, at the same token rates as GPT Image 2. There is no
gpt-image-2.5 alias; name the variant in the request.How do I know whether I have Images 2.5 in ChatGPT?
OpenAI's September 8 announcement says Images 2.5 is rolling out to ChatGPT, ChatGPT Work, and Codex users across all tiers. Check the availability notices shown in your account; image appearance alone does not establish the version. This article has not verified which image variant handles any individual ChatGPT or Codex request. For a repeatable API evaluation, specify
gpt-image-2.5-flare or gpt-image-2.5-sunburst as documented in the OpenAI image generation guide.Should I migrate to GPT Image 2.5 now?
If you call OpenAI directly, Flare at
auto quality is OpenAI's recommended starting point, with Sunburst reserved for editing precision. If you run on EvoLink, choose an explicit quality such as the medium default (auto is not supported on these EvoLink routes), move editing-heavy jobs to Sunburst after a paired test, and keep GPT Image 2 as the frozen baseline for that comparison.Sources and update log
- OpenAI API changelog, September 8, 2026 entry, checked September 8, 2026
- OpenAI model catalog, checked September 8, 2026
- OpenAI GPT-Image-2.5 Sunburst model page, checked September 8, 2026
- OpenAI GPT-Image-2.5 Flare model page, checked September 8, 2026
- OpenAI API pricing, checked September 8, 2026
- OpenAI image generation guide, checked September 8, 2026
- Introducing ChatGPT Images 2.5, September 8, 2026; quoted claims are OpenAI's
- ChatGPT Images 2.5 system card, September 8, 2026
- OpenAI GPT Image 2 model page, checked September 8, 2026
- Introducing ChatGPT Images 2.0, April 21, 2026
- Arena policy on anonymous unreleased models, checked September 4, 2026
- Community reports of the
imagegen25announcement string, September 3, 2026; community-reported only
Update log: September 9, 2026 (later): both routes went live on EvoLink; EvoLink status rows, migration table, next-step links, and FAQ updated. September 9, 2026: corrected the version-identification FAQ and its evidence limits; rechecked the linked release announcement and API guide. September 8, 2026, rewritten for the release: official names, model IDs, what changed, and the migrate-or-wait decision; pre-release signals moved to their own section. September 4, 2026, first published as the GPT Image 2.5 release-status page with the evidence timeline, the
imagegen25 signal, and the wait-or-ship decision.
