AIGC · OpenAI

GPT Image 2.6 Adds SVG Output, 50% Faster Flare, and Keeps Token Prices Flat

By 王若风 · Nov 5, 2026

Read original on juejin.cn ↗ Google Translate ↗ Alt translation

Native SVG output gives design and web teams a direct, editable vector pipeline without a separate vectorization step, and because token prices stayed flat, the faster model costs the same to run. Teams already on 2.5 can test 2.6 with the same prompts and compare output quality, cost, and latency before committing.

Summary

OpenAI released GPT Image 2.6 on November 4, 2026, split into Sunburst and Flare. Both accept text and image input; Sunburst targets detailed composition and multi-turn edits, Flare favors rapid iteration. Official claims put generation speed 50% higher than 2.5, with Flare at low quality around 6 seconds per image. Token pricing is unchanged: $5 per million text tokens, $8 for image input, and $30 for image output. The model adds PNG, JPEG, WebP, and now SVG output, but SVG is restricted to the generation endpoint with n=1, no streaming, and no editing support. A new complexity parameter controls pre-generation planning separately from quality, and OpenAI recommends starting with low complexity for local edits. The release also begins rolling out animated GIF generation in ChatGPT, with adjustable playback speed. The article reviews public comparison images from X and notes that the speed and consistency claims come from official announcements, not independent benchmarks, so migration should be validated on your own workloads.

Takeaways

— Two model IDs: gpt-image-2.6-sunburst and gpt-image-2.6-flare; the product names Sunburst and Flare are not valid API model IDs.

— Official speed claim: generation 50% faster than 2.5; Flare at low quality around 6 seconds, but this is a vendor claim, not a latency guarantee.

— Sunburst is recommended for detailed compositions and multi-turn edits; Flare for quick iteration.

— Token prices are unchanged from 2.5: $5 per million text input tokens, $1.25 cached text, $8 image input, $2 cached image, $30 image output.

— SVG output works only on the images/generations endpoint, requires n=1, cannot stream, and returns static vectors without scripts, external resources, or embedded bitmaps.

— The new complexity parameter controls pre-generation planning depth, independent of quality; start with low complexity for local edits.

— Animated GIF generation is rolling out in ChatGPT, with adjustable playback speed and shareable results.

— Markup-guided editing is available for both supported 2.5 and 2.6 models.

— ChatGPT now lets you save reference images by name for reuse in later chats.

Conclusions

The 50% speed figure and consistency improvements are vendor claims; the article itself notes that actual latency varies and the X comparisons are single, non-blind tests, so independent benchmarking is needed before trusting them in production.

SVG output removes a manual vectorization step, but the article's own example review shows path quality and editability still need human inspection before delivery.

Keeping token prices identical lowers the barrier to migration; the real decision hinges on whether 2.6's output quality and speed hold up on your specific prompts, which the article recommends testing with a fixed set of inputs.

Concepts & terms

complexity

A generation parameter that controls how much pre-generation planning the model does, independent of rendered quality. Higher values suit complex scenes, but OpenAI suggests starting with low complexity for local edits to save time.

SVG output

A vector image output option that preserves paths, shapes, and colors for further editing. In this release it works only on the generation endpoint, requires n=1, cannot stream, and returns static vectors without scripts or embedded bitmaps.

Source: juejin.cn ↗ Google Translate ↗ Backup ↗