OpenAI releases GPT Image 2.6 with SVG and GIF support

New formats for designers

OpenAI officially released GPT Image 2.6 on November 4, 2026, introducing significant updates to its image generation capabilities. One of the most notable additions is the ability to output SVG files, which are vector graphics consisting of paths and shapes rather than pixels. This makes the model much more useful for creating professional assets like logos, icons, and flat illustrations that need to be scaled without losing quality.

In addition to static vectors, the model now supports the creation of animated GIFs directly from text prompts. Users can describe a subject and its motion to create seamless loops, which OpenAI suggests are ideal for brand mascots or reaction memes. The system allows for specific control over the animation, including a frame count ranging from 2 to 16 and custom durations for each frame.

While these new formats are powerful for production, they are currently restricted to new generations. The model does not support exporting SVGs or GIFs from existing images or edits. According to reapi.ai, these features are designed to streamline the initial creation phase of a design project, allowing for a smoother transition into traditional editing software.

Dual-tier API architecture

The new model is structured into two distinct API tiers designed to serve different user needs. The Flare tier is optimized for speed and rapid iteration, capable of returning low-quality images in as little as six seconds. This makes it a preferred choice for creators who need to explore many different layouts or visual directions quickly before committing to a final design.

For more demanding tasks, the Sunburst tier focuses on high-quality output and visual consistency. It is specifically built to keep characters, products, and brand details stable across multiple rounds of edits. This is particularly useful for professional workflows where maintaining a specific brand identity or a recurring character is a requirement for the final product.

Both models share the same core capabilities, including markup-guided edits and the ability to take both text and image inputs. However, they differ primarily in their latency and how they handle detail. Users can choose between these paths based on whether they prioritize the speed of the Flare model or the reliability of the Sunburst model for complex visual work.

Complexity and planning

GPT Image 2.6 introduces a new complexity setting that allows users to dictate how much planning the model performs before it begins rendering. This setting operates independently from the quality slider, which only controls the level of visual detail. The complexity levels are categorized as low, medium, and high, with high being the default setting for most requests.

A low complexity setting is intended for simple, focused edits, such as changing a background or a specific sign within an image. Conversely, the high complexity setting is better suited for dense scenes that contain many different constraints or complex layouts. By adjusting this, users can manage the trade-off between the model's creative planning and the time it takes to generate the final result.

Importantly, OpenAI has not added a separate fee for using higher complexity levels. This gives users the flexibility to push the model to its limits for intricate scenes without increasing their costs. The model uses this planning phase to better understand the spatial relationships and requirements of a prompt before it commits to the final pixels.

Workflow and integration

The release of GPT Image 2.6 is part of a broader update that includes Images 2.6 in ChatGPT and Codex. For developers, the model is available through the reAPI platform, where it can be accessed using standard image generation endpoints. It is billed based on the actual tokens used for each render, ensuring that users only pay for the specific resources consumed by their requests.

The model supports advanced editing features like markup-guided edits, where a user can provide an annotation layer to show exactly where a change should occur. This allows for precise modifications to an original image while using other images as references for style or content. This combination of tools enables a more controlled creative process than simple text-to-image generation alone.

For those looking to test the new capabilities, the model is already live on the reAPI official channel. Users can experiment with the new SVG and GIF outputs by specifying the format in their API calls. The system also includes automated moderation to ensure that all generated content and reference materials meet safety standards before they are delivered to the user.