GPT Images 2.6 Turns a Prompt Into a GIF — and a Still Into SVG

GPT Images 2.6 shipped on November 4, 2026, across ChatGPT and Codex on web, desktop and mobile. Most coverage focused on speed. The two changes that matter more are quieter: it returns an actual animated GIF, and it exports a still as SVG. Both are file formats rather than pictures, and that is what moves the workflow.

What did GPT Images 2.6 actually add?

Capability	Earlier behaviour	Now
Animated output	Stitch frames elsewhere	Native GIF from text or an image, frame editing and speed control
Vector output	Trace by hand or commission a redraw	SVG export from the generation endpoint
Recurring subject	Re-upload every session	Named reference images, reused across chats

The API splits into gpt-image-2.6-flare, returning an image in as little as six seconds, and gpt-image-2.6-sunburst, which trades speed for consistency across edits. Token pricing did not move from 2.5 — $8 per million image input tokens, $30 per million output.

One boundary decides most workflows before they start: the edit endpoint cannot emit GIF or SVG. Only generation can. Anything that begins from an existing file is an edit, and an edit returns a still.

What can you build now that you could not before?

The capability is narrow. The list of things it unlocks is not.

Animated stickers and reaction images. A short loop on a flat background, sized for a chat client. What needed a template tool plus an export queue is now a prompt and a frame check.

Product demos. Six frames showing a lid opening or a switch flipping. Enough for a landing page hero where a video file would be too heavy.

Email and documentation assets. Email clients strip video but render GIF; docs need something that plays without a plugin. This is the gap GIF fills that video never could.

Icon sets as vectors. Export the mark as SVG, drop it into a design system, scale it to a favicon or a billboard without blur. One generated SVG runs about 3 KB against 50 KB as PNG.

Sketches turned into line art. A photograph of a whiteboard becomes scalable art for slides, faster than redrawing it by hand.

Animated vectors. SVG can be driven by CSS or JavaScript; PNG cannot. A vector handed to a developer becomes an interactive element rather than a flat asset.

None of these are new demands. They are old ones that needed a second tool between the prompt and the deliverable, and GPT Images 2.6 deletes that step.

Where does the GIF path break?

Release-day testing found three defects worth planning around.

Size drifts. A 512 × 512 request returned 358 × 358 through the web interface. The API honoured 512. If exact dimensions matter, use the API.

The loop seam shows. A two-second loop returned to its first frame with a visible jump. Frame stepping in the chat preview hides this; a downloaded file does not.

Motion steps unevenly. Ten frames at 200ms do not read as smooth for wide travel. Sparse frames suit small movements, not camera-style sweeps.

A cost surprise sits behind that: one comparison run priced the same still prompt at about $0.0090 on 2.5 and $0.0284 on GPT Images 2.6, roughly three times more, from 439 output tokens to 1,449. A two-second 512px GIF through the API measured about $0.0559 and 17.6 seconds — cheap enough that the frame check, not the spend, is the real cost.

When is SVG export the wrong file?

SVG solves scaling, not production. Model-exported SVG arrives with redundant paths, malformed shapes and substituted glyphs, so every file needs a pass through a vector editor. Print vendors usually want PDF rather than SVG. And text left as a font reference breaks on any machine without that font; text rendered as paths survives.

For longer motion, GIF is the wrong container — a dedicated editor is cheaper per minute. Where that handoff happens is covered in our Concat video editor walkthrough.

What does one GIF cost?

Route	Cost	Notes
API, 2-second 512px GIF	≈ $0.056	10 frames × 200ms, 17.6s to generate
ChatGPT subscription	Included	Subject to plan usage caps
Template subscription	$1.85–$39.50 per asset	No prompt control over motion

For a few assets a month the subscription wins on convenience. For bulk generation or anything embedded in a product, the API is the only route that controls size and timing.

Frequently asked questions

Can GPT Images 2.6 make an animated GIF? Yes, from a text prompt or an uploaded image, with frame-level editing and speed control inside the conversation.

Can it export SVG? Yes, from the generation endpoint. Editing cannot output either format.

Is the GIF the size I asked for? Not always through the web interface — 358 × 358 came back for a 512 × 512 request. The API respected the request.

Is a generated SVG ready to print? No. Expect stray paths and substituted glyphs, and expect the printer to ask for PDF.

Does this replace a video tool? Only for short loops. For longer motion a dedicated editor stays cheaper — see our Tarko AI review for where prompt-driven video stops being enough.

Is output consistent between runs? Better than before. Named reference images pin the subject, which is what makes iterative work possible — the same shape seen in the sticker banner trend.

Sources

• OpenAI on X, November 4, 2026 — availability across web, desktop and mobile
• OpenAI API documentation — generation endpoint, GIF and SVG parameters, edit limits
• kingy.ai, November 4, 2026 — single-run 2.5 versus 2.6 comparison
• freelogoservices.com, sinqlo.com — SVG versus PNG file sizes

---

Published November 5, 2026. Last updated November 5, 2026.