OpenAI's Images 2.6 turns generated images into editable assets. Pick Flare or Sunburst on speed, not price

In brief

OpenAI announced Images 2.6 on 4 November 2026 as two models: GPT-Image-2.6 Sunburst, aimed at detailed visual work, and GPT-Image-2.6 Flare, aimed at fast iteration. The API IDs are gpt-image-2.6-sunburst and gpt-image-2.6-flare, and there is no mini variant. Both are in ChatGPT and Codex on all tiers, and in the API, where Organisation Verification may be required.

OpenAI says Images 2.6 "generates images 50% faster than before". It gives no benchmark or baseline, so treat that as a vendor claim. It also says people now create 4 billion images a week across ChatGPT, Codex and the API, more than half a billion a day.

From regenerate-and-hope to an editable pipeline

The practical change is in the editing tools. Markup edits, already in ChatGPT and now in the API (also documented for GPT Image 2.5, so this is an availability change, not a 2.6-only capability), let a developer send a transparent PNG with sketches to /v1/images/edits and describe the change in the prompt. Markup cannot be combined with a mask. SVG export returns one static vector per generation, with no scripts or embedded rasters, that teams can edit rather than regenerate. Animated GIF output supports 2 to 64 frames (default 9) with frame durations of 20 to 2,000 milliseconds, for generations only and without streaming.

Our reading: for production teams these matter more than raw quality.

Same price, different trade-off

Both models cost the same: per 1 million tokens, $8 for image input, $30 for image output, $2 for cached image input, $5 for text input and $1.25 for cached text input, matching GPT Image 2.5. These are Standard rates: OpenAI lists a Batch tier at half price for GPT Image 2 and 2.5 but not yet for the 2.6 models, so batch workloads moving to 2.6 lose that discount. There is no separate fee for complexity or markup edits. For SVG, the API renders a reference image billed at the image model's $30 per million output rate, then bills the authoring model's tokens used to produce the SVG, so the total depends on how many tokens each model uses.

So the choice is latency versus detail: OpenAI says Flare can return an image in "as little as six seconds" at low quality, while Sunburst offers higher quality and more consistent edits.

The complexity default is a gotcha

A new complexity parameter (low, medium or high) sets how much the model plans before rendering. It defaults to high. OpenAI warns that higher settings can increase latency and introduce unwanted changes, and recommends low complexity for most edits. Set it explicitly in edit workflows. OpenAI's documented limitations also still apply: complex prompts can take up to two minutes, text placement can struggle, and consistency and precise composition are not guaranteed.

Labelling is the publisher's job

ByteNib's house standard is that generated images are labelled as AI-generated wherever they are published. That responsibility sits with the organisation publishing the image, whatever tool produced it.

Sources and scope

Facts are from OpenAI's Images 2.6 announcement, its image generation guide and its API pricing page, all checked on 5 November 2026. The speed and usage figures are OpenAI's own and were not independently verified. The guide's size and quality section and most code examples still name the GPT Image 2.5 models, so resolution and quality options were not verified as 2.6-specific. SVG cost details are from the guide's GPT Image 2.6 costs section, re-checked on the same day because the guide changed during drafting. Early-access customers named by OpenAI include OpenArt, Figma Weave and Town. Workflow interpretation and the labelling guidance are ByteNib editorial analysis.

Drafted with AI assistance and reviewed by the editor before publication.

Continue exploring: AI & Machine Learning analysis, the related implementation tutorial, and the structured learning path.