GPT Image 2.6: What Is Confirmed, What Is Rumor, and What to Build On Today

GPT Image 2.6 is the next name to watch in OpenAI’s image line, and this time the evidence comes from OpenAI itself rather than from screenshots. It is still not a launch. This post separates what can be verified from what is guessed, and answers the practical question: what should you build on right now?

Short version: build on GPT Image 2.5 today, keep the model ID in configuration, and expect 2.6 to drop in on the same request shape.

What is actually confirmed

Everything in this section can be checked against a primary source as of November 1, 2026.

OpenAI’s API spec named GPT Image 2.6 on October 28. A change to OpenAI’s public openai-openapi repository, committed at 20:56 UTC on October 28, added a new markup property to the image edit endpoint. Its description ended with: “Supports GPT Image 2.5 and 2.6. Cannot be combined with mask.” That is the first time “2.6” appeared in an official OpenAI file.

The 2.6 mention was trimmed three days later. A follow-up change on October 31, titled “Clarify markup image support in image editing API”, rewrote the same sentence to “Supports GPT Image 2.5.” The markup property itself stayed. Pulling a name back out of public docs is common when a model is staged but not yet ready to announce, and it is a reason to treat any specific launch date with caution.

There is no GPT Image 2.6 model ID. The spec’s model list still ends with gpt-image-2.5-sunburst and gpt-image-2.5-flare, both with September 8, 2026 snapshots. No gpt-image-2.6 string exists in the model enum, the API changelog, or the official SDKs.

GPT Image 2.5 is the current model. It shipped on September 8 and 9, 2026 in two tiers: Flare, the fast and balanced tier, and Sunburst, the precision tier. Both have text-to-image and edit endpoints, and they are the same models behind ChatGPT Images 2.5.

What markup editing is

The new input is the most concrete thing we know about where OpenAI’s image editing is going. Per the spec, markup is:

• a transparent PNG laid over the first input image to guide the edit;
• the same size as that image after orientation, with transparent areas;
• passed as a file ID, an HTTPS URL, or a base64 data URL;
• not combinable with mask.

The difference from a mask is intent. A mask says where the model may change pixels. Markup is an annotation layer: arrows, circles, scribbled shapes, color blocks or a rough sketch drawn on top of the photo, telling the model what you want there. For a product team that means “circle the logo and write ‘make this blue’” style editing without building your own segmentation step. For designers it means sketching a new window on a building or blocking out where a second subject should stand.

Markup is documented for GPT Image 2.5 today on OpenAI’s API. Whether 2.6 ships with stronger markup understanding is not stated anywhere.

What is inferred, and how confident you should be

The name. Unlike “GPT Image 2.5” back in early September, “2.6” did appear in an OpenAI file, so the label is more than community shorthand. It was also removed, so OpenAI could still ship under a different name or fold the update into ChatGPT first.

The improvements. Community trackers linked several anonymous image models in Arena blind battles to OpenAI in late October. None has official attribution, and nobody has tied a specific checkpoint to 2.6. Early impressions describe an incremental step over 2.5, with hopes for cleaner fine texture and better non-Latin text. These are anecdotes, not benchmarks.

The timing. The last two OpenAI image releases reached the API on launch day, and spec changes have tended to run days to a few weeks ahead of announcements. That points at a release in the coming weeks, but there is no scheduled date and the October 31 edit is a reason not to bet on one.

Pricing. Nothing has been published. Plan against GPT Image 2.5 rates until OpenAI says otherwise.

What this means if you are building an image feature

The mistake to avoid is waiting. Each new OpenAI image model so far has kept the same basic request: a prompt, optional reference images, size and quality settings, and an image back. Work you ship on GPT Image 2.5 carries over.

1. Ship on GPT Image 2.5 now. On WaveSpeedAI the GPT Image 2.5 API page brings together four endpoints: Flare Text-to-Image, Flare Edit, Sunburst Text-to-Image and Sunburst Edit. Edits take up to 16 reference images, with 1K, 2K and 4K output. Current prices are on each model page. If you would rather not write code, try it in the Image Generator.

2. Keep the model ID in configuration. When 2.6 arrives it will appear as new model IDs on the same platform. Switching should be a config change plus a quality review, not a rewrite.

3. Build your evaluation set now. Save the prompts where 2.5 still disappoints you, such as small or non-Latin text, dense textures, hands, and multi-step edits, together with today’s outputs. On launch day you get a same-prompt comparison instead of relying on other people’s screenshots.

4. Design your edit UI for annotations. If your product lets users edit images, an annotation layer (draw, circle, scribble) is where OpenAI’s API is heading. Building the canvas now means you can pass it straight through when markup-guided editing is available on your endpoint.

For choosing between the current tiers, see GPT Image 2.5: Flare vs Sunburst and when to upgrade and the Flare and Sunburst API guide. For how 2.5 compares with its predecessor, read GPT Image 2.5 vs GPT Image 2.

What we will do when it ships

WaveSpeedAI adds new OpenAI image models as endpoints alongside the existing ones, with the same request format, and keeps the previous models available so you can migrate on your own schedule. When GPT Image 2.6 is released, we will update this post with the model IDs, what changed, and a same-prompt comparison against GPT Image 2.5.

Until then, the model to build on is the one you can call: GPT Image 2.5 on WaveSpeedAI.

FAQ

Has OpenAI released GPT Image 2.6?

No. As of November 1, 2026 OpenAI has not announced GPT Image 2.6 and there is no GPT Image 2.6 model ID in the API. The name appeared once, on October 28, in a field description in OpenAI's public API specification.

Where did the name GPT Image 2.6 come from?

From OpenAI's own openai-openapi repository. A change on October 28, 2026 added a markup input to the image edit endpoint with the note 'Supports GPT Image 2.5 and 2.6'. A follow-up change on October 31 shortened the note to 'Supports GPT Image 2.5'.

What is the markup input?

According to the spec, markup is a transparent PNG laid over the first input image to guide an edit. It must match that image's dimensions and contain transparent pixels, and it cannot be combined with a mask. In practice it lets you draw on the image to show where and what to change.

Can I use GPT Image 2.6 on WaveSpeedAI today?

Not yet. The newest OpenAI image model you can call today is GPT Image 2.5, with Flare and Sunburst tiers for text-to-image and editing. See the GPT Image 2.5 API page at https://wavespeed.ai/gpt-image-2-5-api.

Will WaveSpeedAI support GPT Image 2.6 when it ships?

Yes. New OpenAI image models are added as endpoints on the same platform with the same request pattern, so code written against GPT Image 2.5 should only need the model ID changed and a quality check.