OpenAI Images 2.6 Adds GIFs, SVGs and Two API Models

OpenAI released Images 2.6 on November 4, 2026, making the image model family globally available in ChatGPT and Codex across all subscription tiers. Developers also get two API models, GPT-Image-2.6 Sunburst and GPT-Image-2.6 Flare, for image generation and editing. The release adds animated GIF generation, SVG output, markup-guided editing and an adjustable complexity setting. (OpenAI announcement)

The real story is how these pieces fit together. Animation expands the kinds of assets the models can produce, while SVG output and more targeted editing controls address the work of preparing visuals for websites, apps and other production uses. Developers also get an explicit choice between a speed-focused model and one OpenAI positions for higher quality and more consistent edits.

What Is Available Now, and What Is Rolling Out?

OpenAI says Images 2.6 is available globally in ChatGPT and Codex across all subscription tiers, on desktop, mobile and web. That is the broad product-availability statement in its launch announcement. (OpenAI announcement)

The API release is also documented as shipped. OpenAI’s November 4 API changelog lists these model IDs:

• gpt-image-2.6-sunburst
• gpt-image-2.6-flare

The same entry confirms image generation and editing, adjustable complexity, SVG generation and animated GIF generation. It also adds markup-guided editing for supported GPT Image 2.5 and 2.6 models. (API changelog)

ChatGPT’s GIF feature has a narrower rollout caveat. GIF creation is described as rolling out, so the model family’s global availability should not be read as a guarantee that every account has the animation interface simultaneously. The distinction is between a released model family, released API capabilities and the account-level arrival of a particular ChatGPT feature. (OpenAI announcement)

For developers, the practical news is that both API models are released. For ChatGPT users, access to Images 2.6 and access to its GIF-creation interface may arrive on different schedules.

Sunburst vs. Flare: Quality or Faster Iteration

OpenAI gives the two API models different priorities. Sunburst is positioned for detailed work and consistency across edits, particularly at medium or high quality. Flare emphasizes fast generation and editing, with improved layouts for posters, product imagery and other professional visuals. (OpenAI announcement)

OpenAI claims Images 2.6 generates images 50% faster than the previous version. It also says Flare can produce an image in as little as six seconds when quality is set to low. Those are different claims: the first describes a generational improvement, while the second is a conditional latency figure for one model and setting. (OpenAI announcement)

These performance and quality claims are vendor-reported; the supplied sources contain no independent benchmark. The announcement also does not provide enough workload detail to treat six seconds as typical latency or the 50% figure as a guaranteed reduction in every application’s waiting time.

I’d use the distinction as a starting hypothesis for evaluation: Flare for workflows where users repeatedly request alternatives, and Sunburst where preserving details through successive edits is the central requirement. That follows OpenAI’s stated positioning, but a team should measure how often each model produces an acceptable result. A quick first image offers limited benefit if it takes several extra revisions to finish.

Complexity and Quality Control Different Things

The most consequential control may be the separation between complexity and image quality.

OpenAI describes quality as controlling visual detail and fidelity. Complexity adjusts how much planning the model does to satisfy a request, including how much goes into the scene. Its recommendation is to use low complexity for most editing tasks, paired with whatever image quality the workflow requires. Medium or high complexity is suggested for scenes with many elements or constraints. (OpenAI announcement)

That creates a useful combination: a simple edit can still request high visual quality. A product image needing a localized change does not necessarily need the same planning effort as a crowded scene with multiple objects, spatial relationships and layout requirements.

Conversely, a rough composition can be visually undemanding while still requiring substantial planning. A request that specifies where several objects belong may be structurally complicated even if the intended output is only a draft.

My view is that developers should avoid automatically tying complexity to quality in their interfaces. A single “best” setting would hide the distinction OpenAI has introduced. Separate controls, or task-specific defaults, let an application ask for a polished result without always selecting the highest planning effort.

Markup Makes Editing Instructions More Specific

Markup-guided editing brings a capability already available in ChatGPT to the API. OpenAI says developers can use it to pinpoint where specific changes should happen. The changelog also makes clear that this support extends to selected GPT Image 2.5 models, so it is not exclusively a 2.6 feature. (OpenAI announcement, API changelog)

The practical value is reducing ambiguity. In an editor, a user could identify a particular object or area before describing the change, instead of relying entirely on a phrase such as “the item on the left.” That is an application-design implication of spatial guidance, not evidence that every edit will preserve all surrounding content perfectly.

For teams building image-editing tools, markup and complexity solve different problems: markup specifies where to act, while complexity influences the planning effort used to carry out the request.

SVG and GIF Output Change the Deliverable

SVG output gives developers a vector-format option for assets used across websites, apps and print. SVG graphics can scale without the fixed pixel dimensions of a raster image, making the format useful when the same design needs to appear at different sizes. OpenAI explicitly presents SVG export as a way to carry designs into subsequent production work. (OpenAI announcement)

I would still inspect a generated SVG before treating it as a finished design asset. A vector file’s usefulness also depends on its structure and whether it is convenient to edit. The announcement establishes format support, not a guarantee of a particular layer or path organization.

Animated GIF generation adds motion-oriented outputs such as looping characters, product visuals and reaction animations. OpenAI’s launch examples include animation sprite-sheet prompts with six sequential frames, consistent characters and fixed camera positions. Those examples illustrate the importance of specifying continuity and the loop’s intended movement; they are not a documented six-frame limit. (OpenAI announcement)

API Endpoints and Flare Pricing

Flare’s model reference lists text and image inputs, with image output. It supports the dedicated image endpoints:

• v1/images/generations for generation.
• v1/images/edits for editing.

The same reference marks Responses, Chat Completions, Batch and the video endpoint as unsupported for Flare. Developers adding this model to an existing application therefore need to use its supported image routes. GIF output does not make Flare a video-endpoint model. (Flare model documentation)

OpenAI publishes the following Flare token rates:

These are token prices, not fixed per-image prices. Budgeting requires the token usage for the intended requests and outputs. Flare’s speed-focused positioning also should not be read as proof that it is cheaper than Sunburst; a cost comparison needs both models’ rates and comparable workloads. Account rate limits vary by usage tier and organization. (Flare model documentation)

For an initial implementation, I’d choose the model around the task, keep quality and complexity separate, and record both latency and revision count. That evaluates the tradeoff the release actually introduces: how quickly an application reaches a usable asset, with the required detail and consistency.