GPT-Image-2.6: What Changed, Flare vs. Sunburst, Pricing
OpenAI shipped GPT-Image-2.6 on November 4, 2026 — the image model behind "ChatGPT Images 2.6" in the app, and two API models, gpt-image-2.6-flare and gpt-image-2.6-sunburst. Generation is 50% faster, Flare renders in as little as six seconds, and the release adds GIF creation, reusable reference characters, transparent-PNG markup editing, and SVG export.
GPT-Image-2.6 at a glance
| Announced | November 4, 2026 (@OpenAI, @OpenAIDevs) |
| In ChatGPT | All paid tiers + free, on web, desktop, mobile, and Codex |
| API model IDs | gpt-image-2.6-flare, gpt-image-2.6-sunburst |
| Speed | ~50% faster generation; Flare as fast as ~6 seconds |
| New in the app | GIF generation (speed control, frame-by-frame editing), reusable reference images |
| New in the API | Image markup (transparent PNG overlay), SVG export, complexity parameter (defaults to high) |
| API pricing | Text in $5.00 / image in $8.00 / image out $30.00 per 1M tokens — unchanged from 2.5 |
What's new in ChatGPT Images 2.6
Two features landed on the consumer side, per OpenAI's announcement and the community post:
- GIF generation. Turn a prompt or an uploaded image into an animated GIF, adjust animation speed, and edit individual frames before looping or sharing. This is the headline consumer feature — and the one already spawning its own search queries ("gpt image gif generator", "image to gif ai").
- Reference images. Save reusable references of yourself, your pets, or characters once, then reuse them across future generations without re-uploading. Consistency across a set of images used to require prompt gymnastics; now it's a saved asset.
GPT-Image-2.6 vs. 2.5: what actually changed
GPT Image 2.5 (September 8, 2026) introduced the Flare/Sunburst split and the sketch-to-image feature. 2.6 builds on that base:
| GPT Image 2.5 (Sep 2026) | GPT Image 2.6 (Nov 2026) | |
|---|---|---|
| Speed | 50% lower latency vs 2.x | Another ~50% faster; Flare from ~6s |
| GIF generation | — | ✅ prompt or image → GIF, frame editing |
| Reference images | — | ✅ reusable saved references |
| Markup editing | — | ✅ transparent PNG overlay guides edits |
| SVG export | — | ✅ |
| New parameters | — | complexity (default high) |
| Sketch-to-image | ✅ | ✅ |
| Flare / Sunburst variants | ✅ | ✅ (naming carried over) |
| Token prices | $5 / $8 / $30 per 1M | Same |
Flare vs. Sunburst: which GPT-Image-2.6 model to use
The naming started with 2.5 and carries into 2.6 with the same roles:
gpt-image-2.6-flare— built for speed and interactive work. As little as six seconds per generation. Pick it for iteration loops, live editing UIs, and any product where latency is the feature.gpt-image-2.6-sunburst— built for quality and consistency across edits. Pick it for detailed compositions, final renders, and multi-step edit chains where the image needs to stay stable.
Rule of thumb: prototype with Flare, ship with Sunburst. Both cost the same per token, so the choice is purely about latency vs. fidelity.
GPT-Image-2.6 API pricing
From OpenAI's pricing page, per 1M tokens:
| Model | Text input | Image input | Image output | Cached input |
|---|---|---|---|---|
gpt-image-2.6-flare | $5.00 | $8.00 | $30.00 | $2.00 |
gpt-image-2.6-sunburst | $5.00 | $8.00 | $30.00 | $2.00 |
Identical to 2.5, and the 2.6 models don't yet appear in the Batch pricing tables. One caveat worth knowing before you migrate: per-token prices didn't change, but token counts can. Early community testing reports that low/medium quality settings bill noticeably more output tokens on 2.6 than 2.5 for similar prompts (~1,136 tokens on 2.6-medium vs. ~344 on 2.5-medium in one report). Test your real prompts before assuming cost parity, and use the cost calculator in the image generation guide.
Prefer not to set up an API key and do token math? You can generate with GPT-Image-2.6 on Oimi — text to image, image to image, and transparent-background PNGs, with free credits to start.
GPT Image version history (2025–2026)
For anyone keeping score — this cadence is why image-model terms keep surfacing as fresh search keywords:
| Version | API ID | Date | Headline |
|---|---|---|---|
| GPT Image 1 | gpt-image-1 | Apr 2025 (API) | 700M+ images in week one |
| GPT Image 1 Mini | gpt-image-1-mini | Oct 2025 | 80% cheaper |
| GPT Image 1.5 | gpt-image-1.5 | Dec 2025 | 4× faster, cheaper I/O |
| GPT Image 2 | gpt-image-2 | Apr 2026 | Reasoning inside generation |
| GPT Image 2.5 | flare / sunburst | Sep 2026 | Sketch feature, Flare/Sunburst split |
| GPT Image 2.6 | gpt-image-2.6-flare / -sunburst | Nov 2026 | GIFs, references, markup, SVG |
Six model names in nineteen months. Every one of them was a search term before it was a keyword-tool entry.
FAQ
What is GPT-Image-2.6?
OpenAI's image generation and editing model released November 4, 2026, as "ChatGPT Images 2.6" in the app and as gpt-image-2.6-flare / gpt-image-2.6-sunburst in the API. Key additions: GIF generation, reusable reference images, transparent-PNG markup editing, and SVG export, with ~50% faster generation. Or skip the reading and try it free on Oimi.
Flare or Sunburst — which should I use?
Flare for speed and interactive work (from ~6s per image); Sunburst for detail and consistency across multi-step edits. Pricing is identical, so choose on latency vs. fidelity.
How much does GPT-Image-2.6 cost?
$5.00 per 1M text-input tokens, $8.00 per 1M image-input tokens, $30.00 per 1M image-output tokens — same as 2.5. Watch output-token counts on low/medium quality, which early reports say bill higher than on 2.5.
Can GPT-Image-2.6 make GIFs?
Yes — in ChatGPT, from a prompt or an uploaded image, with speed control and frame-by-frame editing. Announced as part of Images 2.6.
Is GPT-Image-2.6 available to everyone?
Yes — all ChatGPT tiers on web, desktop, and mobile, plus Codex and the API, as of November 4, 2026.
The version-history table and launch timeline in this post are sourced from OpenAI's announcements, the OpenAI developer pricing page, the OpenAI community post, and Wikipedia's GPT Image page. Pricing and features move fast — verify against the official pages before committing.