Model update
GPT Image 2.6: What Changed and How to Start
An image-model update is useful when it solves a production problem: slow iterations, a product that changes during editing, or a visual that needs another output format. GPT Image 2.6 brings new options for those decisions.
OpenAI announced Images 2.6 on November 4, 2026. This guide reviews its announcement and API documentation, checked November 5; it is not a hands-on speed or quality benchmark. We explain the access paths, the main controls and a manageable first campaign experiment.
Quick answer: where GPT Image 2.6 is available
OpenAI lists Images 2.6 in ChatGPT and Codex, with GPT Image 2.6 Flare and Sunburst in the API. For a creator, start in ChatGPT; for an application, select the documented API model ID. These are different access paths, with different billing and controls.
Ciyo's Image Generator now lists GPT Image 2.6 Flare and Sunburst next to 2.5, with an Effort control and an Animated GIF format. GPT Image 2.5 Flare stays the default, so select 2.6 in the model menu before you generate.
OpenAI’s launch announcement on X, checked November 5, 2026. This is a vendor announcement, not an independent benchmark or evidence of Ciyo support.
1. Faster iteration changes what you should test
OpenAI reports faster generation for 2.6. Treat its launch speed claims as vendor measurements rather than a deadline for every prompt. Your image size, quality, scene complexity and delivery path can change the waiting time.
For a fictional café campaign, choose one square coffee image and record request-to-usable-output time. Count a rejected image and its revision as part of the task. A fast first image has little value if the cup needs several repairs. Solo creators can start with this small check; agencies should repeat it across their usual briefs.
2. Flare and Sunburst address different priorities
OpenAI positions Flare for speed and Sunburst for demanding detail and edit accuracy. Start with Flare for draft directions; test Sunburst when the acceptance decision depends on product details, precise layout or consistency after edits.
Keep the original reference and acceptance checklist unchanged when switching. In the café example, inspect the cup's handle, rim, material and position. Neither model's positioning guarantees that every detail survives. Save the original rather than repeatedly editing an already degraded result.
3. Quality and complexity control different things
Quality controls rendering detail; complexity controls planning before rendering. They are separate API fields. OpenAI's prompting guide lists low, medium and high complexity, and recommends choosing settings around the task.
A focused background change is a reasonable place to try lower complexity. A crowded composition with several spatial constraints may justify more planning. State the desired picture in the prompt and set available controls in the interface or request. Writing a parameter name into a prompt does not enable an unsupported UI setting.
The supported complexity values are low, medium and high; omitting the field defaults to high. More planning can increase latency and introduce unwanted changes. For a focused background edit, compare low or medium with high while keeping quality fixed; higher complexity is not automatically better.
In Ciyo, this control is called Effort: Low, Medium or High, with High as the default. The credit quote is the same at every Effort level, so you can compare levels without a price change. Ciyo offers Medium and High quality for 2.6.
4. Markup: point to the change without painting on the original
For a focused product edit, keep the original as the first input and put a circle or sketch on a separate transparent PNG of exactly the same oriented dimensions. Send that layer through the Image API's markup field; do not flatten it into the photo or combine it with mask. Additional images are references.
Try changing the cup's backdrop to warm sand while preserving the handle and rim. The annotation shows where to work, not a guarantee that nearby pixels stay unchanged. Compare the result with the original. Ciyo's current text-guided editing workflow does not establish support for this separate API markup field.
In the attached fictional café image, change the background inside the annotated area to warm sand. Keep the ivory cup's outline, handle opening, rim, position and lighting direction. Remove all annotation marks from the result. Do not add text or objects. Review the cup and nearby edges against the original.5. SVG: generate a new vector mark, or trace an existing PNG
For a new simple mark, GPT Image 2.6 supports output_format: svg on the image generation endpoint with n: 1. The API makes a reference image and authors static vector content from it. This output path is not supported by the image edit endpoint. Request paths and flat colors, then inspect the actual downloaded file.
Ciyo already has a different route: select an existing raster logo and use Vector to trace it. Neither tracing nor Make editable guarantees retypable letters. For wording that must remain directly editable, use native canvas Text. Check the new SVG at small sizes, inspect its shapes and colors in a vector editor, and keep the original.
Create one simple icon for a fictional café: a coffee cup with a single rising steam curve. Use only charcoal #242320 and amber #B56A24, with flat fills and a clear silhouette at 24 px. No words, gradients, shadows or photographic texture. Keep the number of shapes modest. Deliver one static SVG through the supported output setting.6. GIF: animate an illustration, preserve a real product with compositing
OpenAI’s direct Images API supports output_format: gif on both generation and edit requests. Use an edit request to supply still reference images. It accepts 2–16 frames at 20–2000 ms per frame, subject to the total pixel budget. Nine frames at 200 ms target a 1.8-second loop. Check the timing in the saved file.
In Ciyo, select a 2.6 model, open Image settings and choose Format: Animated GIF. Ciyo makes one 9-frame loop at 200 ms per frame, 768 px on the long edge, in 1:1, 4:3, 3:4, 16:9 or 9:16. It starts from text only and does not use a reference image. For an exact catalog photo or fixed offer text, the Ciyo compositing workflow keeps the photo and headline static and animates decoration. Check every frame, the last-to-first transition, actual file size and a useful static first frame.
Create a looping illustration for a fictional café. Show one ivory cup on a charcoal background from the first frame. Only a small amber steam curl moves gently upward and returns to its starting shape. Keep the camera and cup still, with no lettering, flashing or scene cuts. Make the last-to-first transition smooth. Target 9 frames at 200 ms per frame using the available GIF controls.Pricing: compare the cost of an accepted result
OpenAI's standard API pricing lists both 2.6 variants at $30 per million image output tokens. Input tokens and other applicable model work can also contribute. This is a token rate, not a fixed per-image quote; equal rates can still produce different task costs.
Record the actual usage for the output size and settings you use. ChatGPT plan access, direct API billing and Ciyo credits are separate pricing systems. In Ciyo, a 2.6 Still image at 1K square costs 6 credits at Medium quality and 7 at High; an Animated GIF costs 6 to 16 credits, set by aspect ratio and quality. Effort does not change the quote. Design agent reasoning is separate Billable LLM usage.
A small migration checklist for your next campaign
Save a successful older-model prompt, its reference and its output. Define the café image's required cup shape, free space for a headline and final crop. Generate with the new model through a supported access path, then make one controlled edit.
Compare the accepted result, number of revisions, complete task time and actual charges. Keep the older workflow when it already meets your requirements. If the new model solves a specific failure, record that reason before expanding to a batch.
Sources and research limits
Sources were checked November 5, 2026. Model capabilities above come from OpenAI documentation; the campaign evaluation recipe is editorial guidance. We have not measured a 2.6 speed gain or compared its outputs in a controlled experiment.
On November 5 we also ran four 2.6 generations in a Ciyo Preview deployment to check settings, saved files and credit charges. That check is not a speed or quality benchmark.
Create with 2.6
Choose your model and format, describe the result, and take the request into Ciyo.
GPT Image 2.6 questions
Has GPT Image 2.6 been released?
Yes. OpenAI announced Images 2.6 on November 4, 2026, including the Flare and Sunburst API models. Check the official access documentation for your account and product.
Is GPT Image 2.6 free?
OpenAI says Images 2.6 is available across ChatGPT subscription tiers. That does not mean unlimited generation or a free API. Check the limits shown for your account; API usage is billed separately.
Can I use GPT Image 2.6 in Ciyo?
Yes. Ciyo's Image Generator lists GPT Image 2.6 Flare and Sunburst, with Effort and an Animated GIF format. GPT Image 2.5 Flare stays the default, so select 2.6 in the model menu and read the quote before you generate.
Does GPT Image 2.6 SVG output turn an existing image into editable text?
No. The documented SVG route generates a new static vector file; it is not image editing or a promise of retypable words. Ciyo Vector traces a raster image, Make editable creates picture layers, and native canvas Text is a separate editable element.
Can I combine markup with a mask?
The documented Image API markup route does not allow combining markup with mask or prompt_rewriter. Keep a separate transparent PNG aligned with the first original image and review the output.
Should every task use high complexity?
High is the default when omitted, but it is not automatically best. Lower complexity can suit a focused edit; higher complexity may help complex layouts while adding latency or unwanted changes. Compare with rendering quality held fixed. In Ciyo, this control is called Effort, and it does not change the credit quote.
Turn your campaign idea into a clear brief
Try GPT Image 2.6 Flare or Sunburst in Ciyo's Image Generator, set Effort, or make an Animated GIF from text. Read the generation quote before you start.