GPT Image 2.6

GPT Image 2.6 is OpenAI's image-generation and editing model family, announced on November 4, 2026. The release appears in ChatGPT as ChatGPT Images 2.6 and in the OpenAI API as GPT-Image-2.6 Flare and GPT-Image-2.6 Sunburst. It follows GPT Image 2.5, with faster generation, revised editing behavior, animated GIF output and additional controls for developers.^[1][2][3]

The API models take text and images as input and produce images. Their documented endpoints are image generation and image editing; their model pages do not list audio, video, fine-tuning, Batch or direct Responses API support. A looping GIF is an image output feature, not support for the Videos API.^[2][3]

Release and model variants

OpenAI announced Images 2.6 for ChatGPT, Codex and the API. Its launch article described availability across subscription tiers, while the Help Center distinguished general image access from the paid-plan availability of Images with thinking and described GIF creation as a rollout.^[1][4]

OpenAI describes Flare as the smaller model and Sunburst as the base model. Its prompting guide recommends testing them against the same prompts, references and settings when choosing between them. These descriptions do not disclose parameter counts or establish an independent quality ranking.^[2][3][5]

The launch announcement says generation is 50% faster than before and that Flare can produce an image in as little as six seconds at low quality. These are OpenAI's claims, not a guaranteed response time. The announcement does not provide a full latency test configuration or distribution.^[1]

ChatGPT features

ChatGPT supports creating an image from a description, uploading a still image for editing, and selecting an area to describe a local change. The editing selection is approximate: changes can extend outside the highlighted area.^[4]

Two features accompanied the November 2026 release:^[4][6]

The Help Center listed ChatGPT Images on all tiers and on web, iOS and Android. Images with thinking was listed for Plus, Pro and Business, with Enterprise and Edu described as forthcoming. Availability statements and rollout status were checked on November 5, 2026.^[4]

API controls and editing

The Image API exposes POST /v1/images/generations for generation and POST /v1/images/edits for editing. Both 2.6 model identifiers appear in the official model catalog and request reference.^[2][3][7][8]

Quality and complexity are separate settings. OpenAI recommends holding one fixed while comparing the other. Higher complexity can add latency or introduce unwanted changes; it is not a guarantee of a better result. Transparent raster output should use PNG or WebP, and the prompting guide recommends inspecting the decoded alpha channel.^[5]

Markup-guided editing uses the original image plus a separate transparent PNG overlay, such as a circle around an object. The overlay must match the first image's dimensions after orientation. It applies to that first image; additional inputs can serve as references. The guide says not to combine markup with mask or prompt_rewriter.^[9]

These controls guide generation rather than guarantee pixel-identical preservation. OpenAI's prompting guide recommends compositing an approved edit into the original when untouched regions must remain exact.^[5]

GIF and SVG output

Animated GIFs

The generation guide documents output_format="gif" on the image-generation endpoint, with frame_count and frame_duration_ms controlling the loop. Both the guide and request reference specify defaults of nine frames and 200 milliseconds per frame. The output is returned as base64-encoded animation bytes.^[7][9]

The documentation had unresolved differences on November 5, 2026. The guide listed 2-64 frames, while the generated request reference limited frame_count to 2-16 and said GIF output required access to a model-specific preview. The guide said the edit endpoint did not support GIF output, although its generated schema included GIF fields. Those differences do not establish that the larger frame range or GIF editing is available to an account.^[7][8][9]

SVGs

SVG generation uses output_format="svg" and n=1. The API first creates a reference image, then an authoring model produces static vector content. Its SVG output excludes scripts, external resources and embedded raster images. The guide does not support SVG output from the image-edit endpoint.^[9]

The request reference describes separate SVG authoring-model selection and reasoning settings, subject to access. This authoring reasoning is distinct from the image model's complexity. SVG generation does not support streaming, partial images, custom output compression or prompt_rewriter.^[7]

API pricing

The following published rates were checked on November 5, 2026. Flare and Sunburst had the same token prices, although the tokens consumed for a particular output can differ.^[2][3][10]

These are token rates, not fixed prices per image. Output size, quality and token consumption affect the bill. The model pages list cached-input prices, but that listing alone does not establish that a particular direct Image API request qualifies for caching.^[2][3][10]

The guide says complexity has no separate fee and markup editing has no additional markup charge. SVG billing includes the reference image and the separate authoring model's token usage. It is therefore not adequately described by the image-output rate alone.^[9]

Safety evaluations and provenance

OpenAI's system card describes checks on requests, input images and generated outputs. In an offline evaluation of 3,077 adversarial requests, the rate of outputs graded unsafe and not blocked was 1.72% for 2.6 Sunburst versus 1.92% for its 2.5 baseline, and 2.14% for 2.6 Flare versus 1.49% for its baseline. Neither overall difference was statistically significant. These results concern the tested moderation configurations, not the frequency of harmful images delivered in production.^[11]

The card notes that GIFs can convey harmful meaning through motion across frames, and that motion-related violations remain difficult to detect. It describes C2PA metadata and SynthID watermarking as complementary provenance signals, without presenting either as a complete provenance solution.^[11]

Limitations

OpenAI's documentation identifies remaining problems with exact text, precise layout and consistency across generations. Its migration guidance calls for checking repeated outputs and complete editing sequences rather than accepting a favorable example as representative.^[5][9]

The release sources provide product documentation, examples and developer-run safety evaluations. They do not establish a universal latency, an independent visual-quality benchmark ranking, or that every announced API feature is enabled for every account. In particular, the GIF schema differences and SVG access requirements matter when implementing those formats.^{[1][7][8][11]}

Continue reading

• OpenAIOpenAI is an American artificial intelligence research and deployment organization.Mentioned 9 times above
• ChatGPTChatGPT is a conversational artificial intelligence chatbot developed by OpenAI and built on the company's GPT series of large language models.Mentioned 8 times above
• OOpenAI APIThe OpenAI API is a REST-based application programming interface that gives developers programmatic access to OpenAI's family of artificial intelligence models, including the GPT series of large language…Linked in the introduction
• GPT Image 2.5GPT Image 2.5 is a family of image generation and editing models released by OpenAI on September 8, 2026.Linked in the introduction
• GPT Image 2GPT Image 2 (API model gpt-image-2, marketed inside the product as ChatGPT Images 2.0) is an image-generation model released by OpenAI on April 21, 2026.Mentioned 3 times above
• SSynthIDSynthID is a family of digital watermarking technologies developed by Google DeepMind for marking and identifying content generated by generative AI systems.Also in Generative AI

More to read

• C2PA (Coalition for Content Provenance and Authenticity)
• OpenAI Codex
• GPT Image 1
• DALL-E
• DALL-E 2
• DALL-E 3

Topics:AI ModelsGenerative AIImage GenerationOpenAI

References

1. ^1 ^2 ^3 ^4OpenAI. Introducing Images 2.6. November 4, 2026.
2. ^1 ^2 ^3 ^4 ^5 ^6OpenAI. GPT-Image-2.6 Flare model. Model, endpoint and pricing documentation, accessed November 5, 2026.
3. ^1 ^2 ^3 ^4 ^5 ^6OpenAI. GPT-Image-2.6 Sunburst model. Model, endpoint and pricing documentation, accessed November 5, 2026.
4. ^1 ^2 ^3 ^4OpenAI Help Center. Images in ChatGPT. Accessed November 5, 2026.
5. ^1 ^2 ^3 ^4OpenAI. Image prompting: GPT Image 2.6. Accessed November 5, 2026.
6. ^OpenAI Help Center. ChatGPT release notes. November 4, 2026 entry, accessed November 5, 2026.
7. ^1 ^2 ^3 ^4 ^5OpenAI. Create image. API request reference, accessed November 5, 2026.
8. ^1 ^2 ^3OpenAI. Edit image. API request reference, accessed November 5, 2026.
9. ^1 ^2 ^3 ^4 ^5 ^6OpenAI. Image generation. Sections on GPT Image 2.6, markup, SVG, GIF, limitations and costs, accessed November 5, 2026.
10. ^1 ^2OpenAI. API pricing. Image-generation rates, accessed November 5, 2026.
11. ^1 ^2 ^3OpenAI. ChatGPT Images 2.6 System Card. November 4, 2026.

Show all 11 references