3AM MarketerOpenAI Images 2.6 Features: 50% Faster Speed and API Specs
OpenAI Images 2.6 cuts generation latency by approximately 50%, introducing dedicated Flare and Sunburst models alongside panel-sliced GIF creation.
OpenAI released Images 2.6 on November 4, 2026, generating visual assets approximately 50% faster than previous versions across ChatGPT, Codex, and the API. OpenAI reports that users create 4 billion images every week across these three environments.12
Alongside faster speeds, the release expands formatting tools with native SVG export, making generated designs scalable for web interfaces and production graphics. For image modifications, Images 2.6 introduces visual markup, enabling creators to point directly to targeted regions of an asset and explain changes in words.13
Flare and Sunburst API model selection
Developers accessing the API have two dedicated model options: GPT-Image-2.6 Flare and GPT-Image-2.6 Sunburst. Model selection balances generation latency against consistency requirements: Flare is tuned for speed, taking as little as six seconds to render an asset at low quality. Conversely, Sunburst prioritizes visual precision and continuity across iterative edits at medium or high quality settings.15
| Model variant | Primary focus | Minimum latency (low quality) | Edit consistency |
|---|---|---|---|
| GPT-Image-2.6 Flare | Speed-optimized generation | 6 seconds | Unspecified (speed priority) |
More on GPT-Image-2.6 Flare
| |||
| GPT-Image-2.6 Sunburst | Detail and precision | Unspecified | High across medium/high settings |
More on GPT-Image-2.6 Sunburst
| |||
API controls for complexity, GIFs, and editing
The release introduces a new complexity parameter that dictates how much planning the model applies before drawing. This parameter operates independently of image quality settings, defaults to high, and accepts three options: low, medium, or high. OpenAI recommends selecting low complexity for standard editing tasks, reserving medium or high settings for scenes containing heavy compositional constraints or crowded subjects.156
Developers face strict operational endpoint boundaries. Output formats such as GIF and SVG are supported exclusively through the generations endpoint, whereas markup editing is isolated strictly to the edits endpoint. Markup workflows use a transparent PNG overlay whose pixel dimensions must precisely match the initial input image.46
Animated GIF creation does not rely on sequential multi-pass rendering. Instead, the architecture paints the entire sequence as a single composite image split into panels before slicing it into individual frames. Frame counts support between 2 and 64 frames, while canvas width and height dimensions must be integer multiples of 16. Individual frame duration can be configured between 20 and 2000 milliseconds, rounded to 10 millisecond increments.457
Watermarking on API images
For digital watermarking, all images created through the official API integrate C2PA metadata alongside SynthID invisible watermarks.58