GPT-Image-2.6 Sunburst and Flare reach OpenAI's API; the API reference says GIF needs preview access and 2.6 model pages list Batch as not supported

Overview

OpenAI added two image models, GPT-Image-2.6 Sunburst and GPT-Image-2.6 Flare, to its API on November 4, 2026, according to the API changelog, which lists adjustable complexity, SVG generation and animated GIF generation, and says markup-guided editing was added for supported GPT-Image-2.5 and GPT-Image-2.6 models. The release accompanies Images 2.6 in ChatGPT, which The Verge reports is rolling out to all ChatGPT and Codex users. OpenAI’s pages differ on GIF access and the maximum GIF frame delay, list Batch support for the 2.5 models but not for 2.6, and a deprecation entry dated 2026-11-10 lists the 2.6 models as the recommended replacement for gpt-image-2. The documentation pages below were read on 2026-11-11; OpenAI’s launch post was read from an Archive.org capture of 2026-11-05 because the live page did not load for the Herald’s fetcher.

What the API adds

OpenAI’s image generation guide says both models support complexity control, markup-guided edits, SVG output and animated GIF output through the Image API.

• Complexity. The complexity field accepts low, medium and high; if it is omitted, both models use high (the API reference says the image model’s configured default applies), and adjusting it has no separate fee, according to the guide. The guide adds that higher complexity can increase latency and introduce unwanted changes. OpenAI’s launch post recommends low complexity for most editing tasks, for faster edits.
• Markup-guided edits. These use POST /v1/images/edits; the annotation layer is sent as markup, has a 50 MB limit, and should not be combined with mask, per the guide. The launch post says image markup, “already available in ChatGPT, comes to the API”.
• SVG. Set output_format to svg on the generations endpoint with n set to 1; the guide says the edit endpoint does not support SVG output and that the returned document is static vector content without scripts, external resources or embedded raster images. The guide adds that billing is the image model’s rate for the reference image plus the authoring model’s rate for the tokens used to produce the SVG.
• GIF. Set output_format to gif on either the generations or edits endpoint. The guide lists frame_count of 2 to 16 (default 9) and frame_duration_ms of 20 to 2000 milliseconds (default 200).

Speed and price

OpenAI’s post says Images 2.6 “generates images 50% faster than before” and that Flare can generate images “in as little as six seconds” when quality is set to low. The post names no baseline model, quality level or complexity setting for the 50% figure. MarkTechPost relays both figures and words the first in its summary box as “than the previous release, per OpenAI”. The Verge reports OpenAI as saying the update enables faster image generation in ChatGPT, and gives no figure. MarkTechPost writes that no benchmark scores or per-image cost table have been published for 2.6. Neither The Verge nor TechRadar reports measured generation times.

On the Sunburst and Flare model pages, image tokens are listed at $8 per million for input, $2 for cached input and $30 for output, and text tokens at $5 for input and $1.25 for cached input. The pricing page lists the same standard-processing rates for the 2.6 and 2.5 models.

Where OpenAI’s pages differ

• GIF and SVG access. The API reference says GIF output “Requires access to the GIF preview for the selected model.” It says the svg object requires access to SVG model selection. The guide, the two model pages, the changelog entry and the launch post, as read, do not mention a preview or an access requirement. The pages do not say which accounts have access.
• GIF frame delay. The guide gives 20 to 2000 milliseconds per frame; the reference gives a minimum of 20 and says GIF output “caps the encoded delay at 655,350 ms per frame”. The Herald does not reconcile the two ranges.
• Batch (the Herald’s comparison of the pages). The 2.6 model pages list Batch as “Not supported”. The GPT-Image-2.5 Flare and Sunburst pages list it as supported and give an output price of $15 per million tokens with Batch processing, and the pricing page’s Batch table has rows for the 2.5 models but none for 2.6. The launch post does not mention Batch.

Deprecations

The deprecations page has an entry dated 2026-11-10 saying gpt-image-2 and the gpt-image-2-2026-04-21 snapshot are deprecated and will be removed from the API on February 10, 2027, with gpt-image-2.6-sunburst or gpt-image-2.6-flare as the recommended replacement. The gpt-image-2 model page shows “Status: Deprecated” and lists Batch as supported. An earlier entry says OpenAI notified developers on June 2, 2026 that gpt-image-1-mini, gpt-image-1.5 and chatgpt-image-latest will be removed on December 1, 2026, and lists the same two 2.6 models as replacements; the pricing page’s Batch table also has rows for those three. The deprecations page does not list the 2.5 models as deprecated.

What the independent coverage shows

The Verge and TechRadar concentrate on ChatGPT. The Verge reports that users can create a GIF from a prompt or an uploaded photo, and TechRadar reports that Images 2.6 reached “all subscription levels, including Free”, and that it brings two new features, GIFs and a persistent reference image. Among the three cited outlets, only MarkTechPost covers the API in detail, and it draws on OpenAI’s own guide and model pages. The API parameters and deprecation dates above are taken from OpenAI’s documentation, MarkTechPost repeats the prices, and none of the three outlets reports a measurement of the speed claims.

What is not established

The sources do not say what the 50% figure is measured against, which accounts can use GIF and SVG output, or whether Batch support will come to the 2.6 models.