OpenAI formally launched Images 2.6 on November 4, 2026, and made it available in ChatGPT, Codex, and the API at the same time. According to the official announcement, the release emphasizes a faster generation cadence and more control over editing rounds. On the same day, the API Changelog recorded the launch of gpt-image-2.6-flare and gpt-image-2.6-sunburst, along with adjustable complexity, SVG generation, animated GIF generation, and markup-guided editing on supported GPT Image 2.5/2.6 models. The launch language is “available globally today,” so this is a live-product analysis rather than a forward-looking preview.

👉 Try the GPT Image 2.6 API on PoYo: per-image pricing and online playground.

Naming requires a clean split between the product layer and the interface layer. ChatGPT and Codex publicly refer to Images 2.6; the API instead calls GPT‑Image‑2.6 Flare (gpt-image-2.6-flare) and GPT‑Image‑2.6 Sunburst (gpt-image-2.6-sunburst). These are not the same field. Production configuration and external commitments should use model IDs, and marketing product names should not be hard-coded.

Speed and the two-tier positioning

According to OpenAI’s product write-up, Images 2.6 image generation is about 50% faster on average than before. In the API, Flare targets fast iteration: when quality is set to low, OpenAI says Flare can finish generation in about six seconds, with improved layout for posters, product shots, and similar professional visuals. Sunburst targets work that needs more detail and precision; OpenAI states that at quality of medium or high, Sunburst can deliver higher-quality generation and keep stronger image consistency across edits. Those percentages and second-scale latency figures come from OpenAI’s announcement. They are not independent third-party benchmarks and should not be rewritten as cross-competitor ranking claims.

The image prompting guide suggests: if an existing GPT Image 2 workflow already meets quality needs, or if you previously used 2.5 Flare, start by evaluating gpt-image-2.6-flare; if existing models fall short on complex tasks, or if you previously used 2.5 Sunburst, start with gpt-image-2.6-sunburst, then compare Flare latency with the same prompts once quality is acceptable. First comparisons should freeze prompts, reference images, size, quality, and output format. Leave complexity at the model default initially, then tune against business acceptance criteria one change at a time.

Capability boundary (as documented)

The product announcement and developer docs jointly list several new or expanded capabilities that define what can be stated externally for this release.

Animated GIF: Generation can emit looping motion via output_format: "gif", suited to brand mascots, product loops, social reactions, and storyboard probes. Official examples and customer quotes frame this as motion expression on the image side. The edits endpoint does not support GIF output, and GIF cannot be combined with streaming or partial images, so it is not a substitute for a full video model.

SVG export: Generation can emit SVG for vector reuse in websites, apps, and print pipelines. Docs note that edits do not support SVG output, SVG requests require n=1, and streaming and output_compression are unsupported.

Image markup: Annotation-based editing that was already available in ChatGPT now enters the API. Developers can use a same-size transparent overlay to pinpoint where changes should happen, supporting product recolors, SKU-local fixes, and similar targeted workflows.

complexity parameter: Controls how much planning the model does before rendering, independent of quality, which controls visual fidelity. When omitted, Flare/Sunburst default to high. OpenAI recommends lower complexity for most edit tasks to speed edits and reduce unnecessary changes; scenes with many elements or constraints can then try medium or high. Changing complexity has no separate charge.

OpenAI says users currently create about 4 billion images per week across ChatGPT, Codex, and the API—more than five hundred million per day. That figure is OpenAI’s published usage scale and should not be extrapolated into third-party quality rankings or competitor scorecards.

Availability, verification, and integration notes

Images 2.6 is available globally across ChatGPT and Codex subscription tiers on desktop, mobile, and web. The API offers Flare and Sunburst; pricing is on each model page. List prices verifiable for this article include $5 per million text input tokens and $30 per million image output tokens, identical across both tiers. Per the image generation guide, organizations may need to complete Organization Verification in the developer console before using GPT Image models. Both Flare and Sunburst model pages list support for v1/images/generations and v1/images/edits. Whether the Responses API can directly target 2.6 models should be judged from the Endpoints table on each model page—both currently mark Responses as unsupported—so do not mix 2.5-era tool configurations into 2.6 production. Safety evaluation and mitigations are covered in the official system card.

Overall, the core Images 2.6 changes are speed, the two-tier choice, and the controllable boundary formed by GIF, SVG, markup, and complexity. Later evaluation should anchor on official model IDs, quality tiers, and output formats, measuring latency, block rates, and per-image cost on business corpora—without writing marketing phrasing into production config or external service commitments.