OpenAI launches GPT-Image-2.6 with GIF support and SVG export

Two models for different needs

OpenAI has officially shipped the GPT-Image-2.6 model family across its API, Codex, and ChatGPT. This update introduces two specific models designed for different user priorities: Flare and Sunburst. While both models share the same underlying architecture, they are optimized for distinct creative workflows.

Flare is specifically built for fast iteration and interactive tasks where speed is the priority. According to OpenAI, this model can return an image in as little as 6 seconds when the quality is set to low. This makes it an ideal choice for developers building applications that require rapid feedback loops or high-volume image generation.

In contrast, Sunburst is designed for high-quality compositions and maintaining consistency across multiple edits. It targets medium or high quality settings to ensure that objects and characters remain stable as the user modifies them. This distinction allows users to choose between a 'fast' mode and a 'precise' mode depending on the final goal of their project.

New formats and complexity controls

One of the most significant technical additions in version 2.6 is the support for SVG and GIF outputs. SVG, or Scalable Vector Graphics, are files that can be resized without losing clarity, making them essential for creating icons, diagrams, and user interface assets. By providing native SVG output, OpenAI is making the tool more useful for professional designers who need crisp graphics.

The update also introduces animated GIF support, which allows users to turn prompts or photos into moving images. In ChatGPT, users can now adjust the speed of these animations and edit individual frames. For social and product teams, this removes the need for a separate animation step, streamlining the content creation pipeline from a single prompt.

Additionally, OpenAI added a complexity parameter to manage how much the model plans before it begins rendering. Users can select from low, medium, or high levels to dictate the density of the scene. Low complexity is recommended for simple edits like changing a background, while high complexity helps with dense layouts or complex relationships between multiple objects.

Markup-guided edits and references

For users who need precise control over specific parts of an image, the new release includes markup-guided edits. This feature allows a user to mark up the exact region of an image they wish to change, rather than relying on text descriptions alone. This type of spatial guidance helps the model understand exactly where a modification should occur without affecting the rest of the composition.

The update also improves how the system handles recurring subjects like people, pets, and characters. ChatGPT users can now save and tag reference images of these subjects to be reused in future prompts. This means the model can maintain a consistent appearance for a specific character without the user having to re-upload the reference image every time they start a new session.

These tools are aimed at reducing the friction of 'prompt engineering' where users often struggle to get the model to stick to a specific visual identity. By combining markup, persistent references, and complexity settings, OpenAI is providing a more professional toolkit for consistent visual storytelling. These features are currently being rolled out to the API for developer use as well as the standard ChatGPT interface.

Performance and pricing details

OpenAI reports that the GPT-Image-2.6 models generate images 50% faster than the previous version. This performance boost is a major update for high-volume users who rely on the API for real-time applications. Despite the speed improvements, the pricing structure remains consistent with the previous version, using a token-based billing system for both Flare and Sunburst.

The cost for image output is set at $30.00 per 1M tokens, while text input costs $5.00 per 1M tokens. Image inputs are priced at $8.00 per 1M tokens, with discounted rates for cached data. While OpenAI has not published a specific per-image cost table for 2.6, these token prices provide the framework for how developers will be billed for their usage.

The demand for these tools continues to grow significantly as the technology becomes more integrated into daily workflows. OpenAI noted that people are now generating more than 4B images per week across their platforms. This is a notable increase from the 3B images per week reported just eight weeks earlier during the launch of version 2.5. This growth highlights the increasing reliance on AI for visual content generation across various industries.

The competitive landscape

While OpenAI offers unique features like native GIF and SVG output, it faces competition from other major players in the image generation space. For example, Google and BFL have published specific per-image prices for their models, such as Nano Banana 2.1 and FLUX 3 Image. These competitors often provide more transparency regarding their exact costs per image, which can be a factor for budget-conscious developers.

OpenAI's current advantage lies in its integration with the ChatGPT ecosystem and its specialized model types for different speeds. However, some competitors offer open-weight versions, like FLUX 3 Image Klein, which can run on consumer-grade hardware like an RTX 3090. This provides an alternative for users who prefer to host their own models rather than relying on a closed API.

Moving forward, the success of GPT-Image-2.6 will likely depend on how well users adopt the new complexity and markup tools. As the market moves toward more complex multi-modal tasks, the ability to generate scalable vectors and consistent characters will be key differentiators. Developers will need to balance the speed of Flare against the high-fidelity consistency of Sunburst to find the best fit for their specific applications.

Next steps for users

Users can begin exploring these new features immediately within the ChatGPT interface or via the OpenAI API. For those looking to experiment with the new complexity settings, it is worth noting that the model defaults to high complexity if no selection is made. This ensures that the highest level of detail is prioritized unless the user specifically requests a simpler, faster render.

Developers interested in the new capabilities should review the specific model cards for Flare and Sunburst. These cards provide the technical details necessary to integrate the new SVG and GIF outputs into existing applications. The guide also clarifies how to use markup-guided edits to achieve more precise results than standard text prompts allow.

As the platform continues to scale, keep an eye on how OpenAI handles the trade-offs between complexity and latency. While higher complexity produces more detailed results, OpenAI warns that it can also introduce unwanted changes or slower response times. Finding the 'sweet spot' between these settings will be a primary task for creators and engineers alike as they integrate GPT-Image-2.6 into their products.