OpenAI Images 2.6 Adds GIFs, SVG Output and a Separate Complexity Control

Images 2.6 separates planning complexity from image quality. Compare Flare and Sunburst pricing, new output formats, and what to test before migrating.

OpenAI launched Images 2.6 on November 4 in ChatGPT, Codex and the API. The release adds GIF creation, SVG exports and markup-guided API edits. Flare targets fast iteration; Sunburst targets more detailed work and consistency across edits. OpenAI’s announcement

For developers, the most useful change to investigate is the separation between complexity and quality. A request to replace one word on a sign should be evaluated differently from a request to compose a crowded technical diagram.

Complexity is a planning control

The image-generation guide says complexity accepts low, medium or high, defaults to high for both models, and has no separate fee. Quality controls rendering quality. Higher complexity can increase latency and introduce unwanted changes, so a higher setting is not automatically preferable for a focused edit.

OpenAI recommends low complexity for most edits in its launch post. Treat that as a starting setting to test. For a narrow change, compare whether the requested region improves and whether everything outside it stays acceptable. For a dense composition, assess relationships between objects as well as surface detail.

The guide now includes a 2.6 section, but its existing examples still use 2.5. Check the model identifier when copying an example rather than assuming the page’s newest feature applies to every sample unchanged. Developer guide

Flare and Sunburst share the listed token rates

The model pages list the following rates for both gpt-image-2.6-flare and gpt-image-2.6-sunburst, per million tokens:

Token categoryInputCached inputOutput
Text$5.00$1.25—
Image$8.00$2.00$30.00

Sources: Flare model page and Sunburst model page.

Equal rates do not establish equal cost per finished image. Compare actual usage and the number of attempts needed at your chosen settings. Save the request configuration with each result so that a faster setting can be reproduced.

OpenAI reports 50% faster generation and says low-quality Flare can produce an image in as little as six seconds. These are vendor claims, not timings measured by S5 Labs. They should not become an application latency guarantee. Launch performance claims

New formats need their own acceptance tests

The developer guide confirms SVG and animated GIF output through the Image API. Test the exact output format on the endpoint your application uses before wiring it into a production export flow. Supported features

For an SVG, inspect scaling, text rendering and whether the result is useful in the destination editor. For a GIF, inspect the loop boundary, frame consistency, file size and readability at the actual display size. These are proposed checks, not results from our own model tests.

Keep a static alternative for motion-based assets on a website. A moving version should add something useful without making essential information dependent on animation.

A migration trial worth running

Start with a small set of existing Images 2.5 workflows: a product image, a text-heavy layout and a narrowly scoped revision. Compare both models at explicit complexity and quality settings. Record elapsed time, usage, rejection reason and whether a reviewer accepts the output.

Update, November 12, 2026: OpenAI’s deprecations page now lists gpt-image-2 and the gpt-image-2-2026-04-21 snapshot as deprecated, with removal from the API on February 10, 2027. It names gpt-image-2.6-sunburst and gpt-image-2.6-flare as the replacements, so applications still calling gpt-image-2 have a fixed date by which to finish a migration trial like the one above.

For workflows that then turn an approved still into video, evaluate Omni 1.2 Flash separately. An image that passes review is a source asset, not proof that the resulting video preserves it.

Updates and corrections

  • Update

    OpenAI's deprecations page now lists gpt-image-2 for removal from the API on February 10, 2027, with the 2.6 models as replacements.

Found an error? Report it through the contact page.

Continue reading.

Insight4 min read

GPT-Live-1 Comes to the API: Voice and Reasoning Get Separate Jobs

GPT-Live-1 brings full-duplex voice to the API. How delegation works, what the per-minute price includes, and what to test before switching.

Insight10 min read

OpenAI Opens a Free CSAM Image Classifier to Vetted Organizations on the Moderation API

OpenAI's CSAM image classifier is free to approved organizations on v1/moderations. What it returns, how access works, measured limits, reporting duties.

Insight22 min read

GPT-6.2 Astra Drops to $8/$40, Sol Holds at $2/$10 and Cyber Safeguards Tighten

GPT-6.2 Astra costs $8/$40 per million tokens, 20% below GPT-6 Astra; Sol stays at $2/$10. Migration steps, benchmark caveats and tighter cyber refusals.