On this page
OpenAI published Introducing Images 2.6 on 2026-11-04, describing a product and API update that lands in ChatGPT, Codex, and the API. The same post says Images 2.6 is available globally that day in ChatGPT and Codex across all subscription tiers on desktop, mobile, and web, and that image generation is about 50% faster than before. DigIntoAI has not timed generations or compared billing against those claims—figures below follow the OpenAI blog, the @OpenAI public post, developer model pages, the API changelog, pricing tables, and the Images 2.6 system card.
The same-day @OpenAI post (https://x.com/OpenAI/status/2118124944928432327) carries created_at 2026-11-04T23:35:14.000Z, which is 2026-11-05 07:35:14 Asia/Shanghai. Its wording is short and useful as a product pin: Images upgraded as ChatGPT Images 2.6, now with GIFs and reference images, plus two new API models and the same 50% faster claim. The blog leads on GIFs, faster iteration, and edit consistency; the public post is the clearest official line that reference images ship with this cut. Both sources are OpenAI self-reports, not DigIntoAI lab runs.
A nearby ChatGPT image product story on this site—OpenAI to Test Visual Ads Inside ChatGPT Image Generation—covers ad testing inside image generation, a different theme from Images 2.6’s GIF / reference-image / dual-API launch.
Where Images 2.6 is available
Per the Introducing Images 2.6 blog:
- Surfaces: ChatGPT, Codex, and the API.
- ChatGPT / Codex: available globally today across all subscription tiers on desktop, mobile, and web.
- Speed (self-report): Images 2.6 generates images 50% faster than before.
- Volume (company self-report, not DigIntoAI stats): people create about 4 billion images per week across ChatGPT, Codex, and the API—on average more than half a billion per day.
“All subscription tiers” is the blog’s wording. For how Plus and Pro seats differ on price and limits in general—not Images 2.6-specific quotas—see ChatGPT Plus vs Pro—Price, Limits, and Who Should Upgrade (November 2026). DigIntoAI has not checked tier UI surfaces for GIF or reference-image controls beyond the blog’s all-tiers / all-clients claim.
GIFs and reference images
GIFs. The blog frames GIF output as motion for brand mascots, characters, product loops, social reactions, and animated storyboards before video work. Example prompts on the post ask for sprite-sheet style multi-frame loops (for example six frames in a 3×2 grid) that read as seamless looping GIFs. Early-access quotes on the post (OpenArt; Figma Weave) describe social-native motion and creative-production workflows; those are partner statements republished by OpenAI, not DigIntoAI tests.
GIF ≠ long video. This launch is still image generation and editing—animated short loops and multi-frame sheets—not a Sora-class long-video model and not a substitute for video APIs. OpenAI’s blog and public post describe GIF / animation capability; they do not describe an adjustable playback-speed control or a frame-by-frame editor for this cut.
Reference images. The @OpenAI post explicitly pairs GIFs with reference images. The blog’s product story stresses keeping characters, products, and brand details consistent across edits, and notes that even a simple sketch can become an expressive animated character. Those lines read as consistency / reference-driven editing language from OpenAI. Official text does not lock a separately branded “personal asset library” product name.
Flare versus Sunburst on the API
Developers get two model IDs, both documented on OpenAI’s developer model pages and listed on the 2026-11-04 API changelog:
| Role (blog / model-page framing) | Model ID | Quality emphasis (OpenAI wording) |
|---|---|---|
| Fast iteration | gpt-image-2.6-flare | When quality=low, Flare can generate in as little as about six seconds; blog also notes better layout for posters and product imagery |
| Detailed visual work | gpt-image-2.6-sunburst | Medium / high quality; higher-quality generation and stronger consistency across edits |
Model-page blurbs match that split: Flare is “image generation and editing for fast iteration”; Sunburst is “for detailed visual work.” Both accept text + image inputs and output images. Supported endpoints on both pages: v1/images/generations and v1/images/edits. Chat Completions, Responses, Batch, Videos, and several other routes are listed as not supported for these image model IDs on the same pages.
Changelog wording for Nov 4: released GPT-Image-2.6 Sunburst and Flare for image generation and editing, with adjustable complexity, SVG generation, and animated GIF generation; added markup-guided editing for supported GPT Image 2.5 and 2.6 models. DigIntoAI has not run either ID on generations or edits, timed the six-second Flare claim, or compared Sunburst edit consistency against Flare on a fixed prompt set.
API controls: markup, SVG, complexity, quality
Beyond the two IDs, the blog and changelog highlight controls meant for multi-step creative pipelines:
- Image markup — already available in ChatGPT; now coming to the API so teams can mark where specific changes should land.
- SVG export — designs ready for websites, apps, and print workflows (API-side export as described by OpenAI).
complexity— new parameter with low / medium / high planning effort. Blog guidance: for most editing tasks, prefer low complexity for faster edits, paired with whatever image quality the workflow needs; for scenes with many elements or constraints, try medium or high.- Quality — still the visual-detail dial (low / medium / high in the Flare / Sunburst framing above). Complexity guides how much planning goes into the scene; quality guides fidelity.
The Images 2.6 system card (published 2026-11-04) similarly notes GIF generation, photo markup in ChatGPT and the API, SVG output in the API, Flare optimized for faster generation, and configurable reasoning / complexity. DigIntoAI has not exercised markup-guided edits, SVG exports, or complexity presets on a live key.
Same token price for Flare and Sunburst
Flare and Sunburst list the same per-million-token rates on their model pages and on the developer pricing table’s image-generation rows. DigIntoAI has not seen an official line that Flare is cheaper or Sunburst is a premium surcharge—same numbers for both IDs:
| Metric | Price per 1M tokens (USD) |
|---|---|
| Text input | $5 |
| Text cached input | $1.25 |
| Image input | $8 |
| Image cached input | $2 |
| Image output | $30 |
Those figures sit in the same token-price band as gpt-image-2.5-* and gpt-image-2 image rows on the pricing table DigIntoAI checked while writing—OpenAI does not describe Images 2.6 as a cut-price or price-hike event relative to 2.5 / 2 in the sources above. Those figures are the developer list prices for these IDs on the pages DigIntoAI checked while writing; pricing tables can change.
For the broader OpenAI API wall beyond these two image IDs, see OpenAI API Pricing (November 2026). This Images 2.6 page carries the Flare / Sunburst rows for search intent beside that broader guide.
Safety system card (one beat)
OpenAI published a ChatGPT Images 2.6 system card on 2026-11-04 covering GIF risks, safeguards, and evaluations. Short takeaways relevant to this product brief: the core safety stack builds on Images 2.5, with additional GIF safeguards; motion-related violations remain hard to catch completely; Flare is framed as the faster path; markup / SVG / complexity appear as product capabilities alongside the safety discussion. Evaluation percentage tables stay on the system card (offline adversarial scores, not production rates). This brief does not claim the stack is “fully safe.”
DigIntoAI status
DigIntoAI has not generated stills or GIFs in ChatGPT or Codex after Images 2.6, has not attached reference images on a live account, has not called gpt-image-2.6-flare or gpt-image-2.6-sunburst on generations or edits, and has not timed the ~six-second or ~50% claims or reconciled a sample invoice against the $5 / $1.25 / $8 / $2 / $30 schedule. Figures above follow OpenAI’s blog, public post, model pages, changelog, pricing table, and system card as of write-up (2026-11-05, Asia/Shanghai).
How this sits next to other image tools
Images 2.6 is an OpenAI ChatGPT / Codex / API image cut. A different vendor’s in-editor image path is summarized in Generate and Edit Images Directly in Copilot—a cross-vendor image-generation pointer beside this OpenAI cut.
Quick facts
| Item | Official pin (self-report) |
|---|---|
| Blog date | 2026-11-04 — Introducing Images 2.6 |
| ChatGPT / Codex availability | Global that day; all subscription tiers; desktop / mobile / web |
| Speed claim | ~50% faster than before |
| Product pins | GIFs + reference images (@OpenAI post 07:35:14 Asia/Shanghai) |
| API IDs | gpt-image-2.6-flare · gpt-image-2.6-sunburst |
| Flare latency claim | quality=low → as little as ~six seconds |
| Sunburst emphasis | medium / high quality; stronger edit consistency |
| Shared $/1M | text $5 / $1.25 cached; image in $8 / $2 cached; image out $30 |
| Extra API controls | markup · SVG · complexity low/medium/high |
| Endpoints | v1/images/generations · v1/images/edits |
| DigIntoAI hands-on | No |
In short
Images 2.6 is OpenAI’s 2026-11-04 image update for ChatGPT and Codex on every subscription tier, plus two API models that share one token-price schedule. Product messaging centers faster still generation, GIF motion, and reference-image / consistency editing; API messaging adds Flare versus Sunburst, markup, SVG, and complexity. DigIntoAI has not hands-on tested any of those surfaces; the numbers and feature pins above are OpenAI self-reports.