What is GPT Image 2.6 (Flare and Sunburst)?
GPT Image 2.6 is OpenAI's image generation and editing family, released on 4 November 2026, two months after GPT Image 2.5. Like 2.5 it comes as two API models: gpt-image-2.6-flare for speed and layout work such as posters and product shots, and gpt-image-2.6-sunburst for detailed images and consistency across chains of edits. The same update powers ChatGPT Images 2.6 on every ChatGPT and Codex tier.
OpenAI says it generates images 50% faster than before, with Flare producing an image in as little as six seconds at low quality. New API controls include a complexity setting (low, medium or high, high by default) that sets how much the model plans before drawing, markup-guided edits from a transparent PNG drawn over the source image, SVG output and animated GIFs. ChatGPT adds saved reference images of people, pets and characters.
How much does GPT Image 2.6 (Flare and Sunburst) cost?
| Plan | Price | What you get |
|---|---|---|
| Image tokens | $8 input ($2 cached) / $30 output per 1M | |
| Text input | $5 per 1M ($1.25 cached) |
Same rates for Flare and Sunburst, per 1M tokens: image input $8 ($2 cached), image output $30, text input $5 ($1.25 cached). Complexity has no separate fee. No Batch pricing, unlike GPT Image 2.5.
Is GPT Image 2.6 (Flare and Sunburst) worth it?
GPT Image 2.6 is a speed and workflow release, not a quality leap. OpenAI claims a 50% cut in generation time and publishes no new quality benchmark; two days after launch it is not yet ranked on Artificial Analysis's text-to-image arena, where its predecessors GPT Image 2.5 Sunburst and Flare hold first and second place at 1227 and 1208.
The speed matters because 2.5 was slow. Flare can now return a low-quality image in about six seconds, which makes interactive design tools and live previews practical. Sunburst gets no speed figure from OpenAI, only better consistency across repeated edits at medium or high quality.
The new controls are aimed at production work. Complexity tells the model how much to plan before drawing: low for a background swap, high for dense text and layouts, at no extra charge, though OpenAI warns higher settings can add latency and unwanted changes. Markup edits let a user circle what should change on a transparent layer instead of describing it in words, within limits of 16,777,216 pixels per image and 50 MB for the markup file. SVG output gives designers vector files from the generation endpoint, and GIF output arrives in both ChatGPT and the API.
Prices did not move: $8 per million image input tokens and $30 per million output, the same for both models. What changed is the endpoint list. The model pages list only image generation and image edits, and mark Batch unsupported, so the 50% discount 2.5 offered on Batch is gone for anyone who upgrades. At max quality Artificial Analysis prices a 1024x1024 image from 2.5 at about $0.21, against $0.034 for Nano Banana 2.1.
Compliance needs a check. OpenAI's data controls page lists the GPT Image 2.5 models, GPT Image 2 and older models as compatible with zero data retention on the image endpoints, and does not yet list 2.6. API inputs are not used for training by default, and image endpoints keep abuse-monitoring logs for 30 days.
Where it earns its place
A front-end engineer building a poster editor for a print-on-demand shop
Show customers a live preview while they type
Flare at low quality and low complexity can return a draft in about six seconds, then a final render at higher quality once the layout is chosen. Token prices match 2.5, so the cost per final poster does not change.
A creative operations lead at a 40-person e-commerce agency
Let account managers request edits without writing long prompts
Account managers circle the product or background to change on a transparent markup layer and send it to the edits endpoint with Sunburst for consistency across rounds. Each file must stay under 16,777,216 pixels.
A marketplace engineer rendering 50,000 catalogue images a month
Keep the monthly image bill predictable
Stay on GPT Image 2.5 through Batch for overnight jobs, which halves token rates, and use 2.6 Flare only for interactive requests. 2.6 has no Batch support.
Before you commit
- Setup
- Call v1/images/generations or v1/images/edits with model set to gpt-image-2.6-flare or gpt-image-2.6-sunburst; add complexity, and a markup PNG on edits, as needed.
- Integrations
- Image API generation and edit endpoints only; Batch, Chat Completions and Responses are listed as unsupported on the model pages. Also served through fal.
- Data and privacy
- API inputs and outputs are not used for training by default. Image endpoints keep abuse-monitoring logs for 30 days; OpenAI's zero data retention list for image endpoints names the 2.5 models but not 2.6, and flagged image inputs are kept for review even under ZDR.
- Cost at scale
- No per-image price is published for 2.6; at 2.5's comparable max setting a 1024x1024 image costs about $0.21, so 10,000 images run to roughly $2,100 with no Batch discount available.
- Limits
- Markup files up to 50 MB and images up to 16,777,216 pixels; markup and mask cannot be combined; higher complexity can add latency and unwanted changes.
What’s good
- OpenAI claims 50% faster generation; Flare can return an image in about six seconds
- Complexity control at no extra charge
- Markup-guided edits from a transparent overlay
- SVG and animated GIF output through the Image API
- Unchanged token prices from GPT Image 2.5
What’s not
- No Batch support, so no 50% bulk discount
- Not yet listed as zero data retention compatible
- No independent quality ranking yet
- Still several times dearer per image than Nano Banana 2.1 or MAI-Image-2.6
- Not callable directly on the Responses endpoint
Best for
Product and design teams on OpenAI's API who need faster interactive image generation and precise, pointed edits.
GPT Image 2.6 (Flare and Sunburst) alternatives
GPT Image 2.6 (Flare and Sunburst): common questions
- Is GPT Image 2.6 (Flare and Sunburst) worth it?
- A real speed upgrade on the strongest image family available, but teams that relied on 2.5's Batch discount or zero data retention should not switch yet. We rate it 4.5 out of 5.
- How much does GPT Image 2.6 (Flare and Sunburst) cost?
- Same rates for Flare and Sunburst, per 1M tokens: image input $8 ($2 cached), image output $30, text input $5 ($1.25 cached). Complexity has no separate fee. No Batch pricing, unlike GPT Image 2.5.
- What are the downsides of GPT Image 2.6 (Flare and Sunburst)?
- No Batch support, so no 50% bulk discount; not yet listed as zero data retention compatible; no independent quality ranking yet; still several times dearer per image than Nano Banana 2.1 or MAI-Image-2.6; not callable directly on the Responses endpoint.
- Who is GPT Image 2.6 (Flare and Sunburst) best for?
- Product and design teams on OpenAI's API who need faster interactive image generation and precise, pointed edits.
- How much does GPT Image 2.6 (Flare and Sunburst) cost for a team or in production?
- No per-image price is published for 2.6; at 2.5's comparable max setting a 1024x1024 image costs about $0.21, so 10,000 images run to roughly $2,100 with no Batch discount available.
- Is GPT Image 2.6 (Flare and Sunburst) safe for company data?
- API inputs and outputs are not used for training by default. Image endpoints keep abuse-monitoring logs for 30 days; OpenAI's zero data retention list for image endpoints names the 2.5 models but not 2.6, and flagged image inputs are kept for review even under ZDR.
- What are the best GPT Image 2.6 (Flare and Sunburst) alternatives?
- The closest alternatives we have reviewed are GPT Image 2.5 (Flare and Sunburst), Nano Banana 2.1, MAI-Image-2.6 and Grok Imagine Image 2.0.
Answers checked Nov 6, 2026.
Make GPT Image 2.6 (Flare and Sunburst)? Show this rating on your site
