OpenAI Releases GPT-Image-2.6: Flare and Sunburst Dual Models Generate GIF/SVG in 6 Seconds

OpenAI has made another move in the image generation space. According to MarkTechPost, OpenAI has officially released Images 2.6 along with two API models, GPT-Image-2.6 Flare and Sunburst. This update doesn't stop at image quality tweaks—it focuses on three engineering dimensions: output formats, editing capabilities, and generation speed.

Flare Leads on Speed: Fastest 6-Second Image Generation via API

The first difference mentioned between the two models is speed. Flare can complete an image generation in as fast as 6 seconds, a figure aimed at API call scenarios. For products that embed image generation into workflows, latency often affects experience more than the precision of a single image—users start doubting whether a request failed once they wait more than ten seconds.

A 6-second generation time means Flare is better suited to batch, interactive calls—scenarios like e-commerce asset previews or rapid iteration on marketing images that require "tweak and see." OpenAI didn't specify in the materials whether Sunburst makes trade-offs on speed, but given the dual-model parallel naming, the two are most likely aimed at different load profiles.

Output Format Expansion: Native GIF and SVG Support

Images 2.6 fills two obvious gaps at the output level.

GIF Output

Previously, most image models only produced static bitmaps, while animations required extra stitching or going through video models. Native GIF support lets the model directly generate looping animation assets—a substantial convenience for social media content and lightweight motion displays.

GIF is a bitmap animation format born in 1987 that achieves motion by chaining multiple frames within a single file and looping playback. Its technical advantage is extremely broad compatibility—almost all browsers, social platforms, and chat apps support it natively without plugins or additional decoding; it also supports simple transparency and a 256-color palette with a simple file structure. The cost is limited color representation, compression efficiency inferior to modern video codecs, and bulky file sizes for long animations. In recent years, WebP, AVIF, and short video formats have surpassed it in both quality and size, but GIF remains hard to fully replace in memes, lightweight motion, and cross-platform sharing scenarios—which is the practical reason image models choose native GIF support rather than only outputting video.

SVG Export

SVG's significance is more production-oriented than GIF. Vector formats can scale losslessly and go directly into design tools and front-end code, avoiding the broken chain of "generate bitmap then manually trace." For assets like icons, illustrations, and UI elements that need multi-size reuse, SVG export turns AI-generated results from "reference images" into "directly deliverable assets."

Together, these two capabilities show OpenAI is pushing image models toward "a link in the design pipeline" rather than just toys in a chat window.

SVG (Scalable Vector Graphics) is an XML-based vector image format that describes images using mathematical paths, geometric shapes, and text instructions, rather than recording color values per pixel grid like JPEG and PNG. Because of this, SVG images never develop jagged edges or blur no matter how much they're scaled, and file sizes are typically far smaller than high-resolution bitmaps of equivalent visual quality. For front-end developers, SVG can be embedded directly in HTML, styled with CSS, and dynamically modified via DOM operations, making it naturally suited to responsive layouts and multi-theme scenarios. But having AI models directly output usable SVG isn't easy: it requires the model to understand the structural composition of graphics—which parts are paths, which are fills, how hierarchy is organized—rather than just predicting pixel distributions. This is precisely why native SVG export has advantages in quality and editability over "generating bitmaps then vectorizing."

Markup Edits and Complexity Parameter: More Control

Beyond formats, Images 2.6 also introduces markup edits and a complexity parameter.

Markup edits allow users to make local annotations on images to guide modifications, which is closer to designers' actual working habits than pure text prompts—circling the area to change rather than describing in a paragraph "make that thing on the left a bit bigger." The complexity parameter gives callers a tuning knob to trade off between generation quality and resource consumption—a practical control for cost-sensitive applications.

Interestingly, both capabilities lean toward "process control" rather than "result beautification," reflecting OpenAI's judgment on the image API's positioning: users want more than a good-looking image—they want one that can be repeatedly adjusted and plugged into pipelines.

What This Update Means

From public information, the core changes in the GPT-Image-2.6 series center on the engineering side: lower latency, more output formats, finer editing control. Updates like these won't flood social media, but they have a more direct impact on teams actually building image generation into products.

Native GIF and SVG support deserve particular attention. When models can directly output vector files and animations, downstream format conversion and manual patching are reduced, and the "usability rate" of AI-generated content rises accordingly. Flare's 6-second generation lowers the barrier for real-time interaction another notch.

It should be noted that the material information is fairly limited this time. OpenAI hasn't yet disclosed in reports the specific pricing of the two models, resolution caps, the specific value ranges of the complexity parameter, or how Sunburst differs in positioning from Flare. These details directly affect developers' selection decisions and are worth waiting for official documentation to clarify.

For teams already using image APIs, two things are worth evaluating first: the fidelity of SVG export in actual design delivery, and whether markup edits can replace some manual rework in multi-person collaboration workflows. If these two meet expectations, the range of viable use cases for image generation will broaden considerably.

Designs like the "complexity parameter" are gradually becoming common practice in generative AI APIs, borrowing ideas from adjustment options in image generation such as sampling steps and guidance scale: callers use a scalar or tier to trade off between output quality, generation latency, and compute cost. For cost-sensitive scenarios like batch calls and preview iteration, lowering complexity can significantly reduce unit price and wait time; for final delivery stages, raising it can yield more refined results. The key to understanding this knob is that it hands the decision-making power over "generation budget" to the application layer, enabling the same model to cover different stages from draft to finished product, rather than applying fixed configurations to all requests.

One-line Summary

OpenAI releases Images 2.6 and two API models, Flare and Sunburst, focusing on formats, editing, and speed.