Money and business in the Middle East.

AI

OpenAI launches Images 2.6 with animated GIFs, 50% faster generation

Animated GIFs generated from a text prompt or an uploaded image are now part of OpenAI’s image stack, launched November 4, 2026 as Images 2.6 across ChatGPT, Codex and the API. OpenAI’s own summary describes faster generation, sharper editing and new API controls.

· Originally published by ontime+ · Last verified: 5 Nov 2026 (Nada Salam)

Key Points

  1. OpenAI released Images 2.6 on November 4, 2026, adding GIF creation and two new API models.
  2. The company says generation runs about 50% faster, with Flare producing low-quality images in roughly 6 seconds.
  3. Marketing and product teams can now build animation, precise edits and print-ready SVGs without third-party tools.

The latest:

Animated GIFs generated from a text prompt or an uploaded image are now part of OpenAI’s image stack, launched November 4, 2026 as Images 2.6 across ChatGPT, Codex and the API. OpenAI’s own summary describes faster generation, sharper editing and new API controls. The company says image output across its products exceeds 4 billion images a week.

Details:

  • The speed claim: OpenAI says generation is roughly 50% faster than the previous release, according to TestingCatalog and OrcaRouter. The new Flare model returns an image in about 6 seconds at low quality, the same reports said, a tier aimed at rapid iteration rather than final-grade output.
  • GIF controls: According to The Verge via unite.ai, users can generate a short animated GIF from a prompt or an uploaded image, then adjust playback speed, review individual frames and share the result. OpenAI’s demonstration featured a dancing cat.
  • GIF use cases: TestingCatalog lists characters, mascots, product shots, diagrams, reaction clips and storyboards as intended applications. OrcaRouter noted a one-way limit: the API can output an animated GIF but will not accept a GIF as an input file.
  • The two models: The API gains GPT-Image-2.6 Flare, tuned for speed and professional layouts such as posters and product images, and GPT-Image-2.6 Sunburst for finer detail at medium and high quality. OrcaRouter identified the second as gpt-image-2.6-sunburst from documentation.
  • Edit consistency: OpenAI Developers said that at medium or high quality, “Sunburst maintains detail consistency from one edit to the next,” per TestingCatalog, which said characters, products and brand details hold across successive rounds. Users can also store reference images of themselves, their pets or their characters for reuse without repeated uploads.
  • The markup tool: ChatGPT’s region-selection tool now reaches the API, letting developers mark exactly where a change applies, OrcaRouter reported. It is sent as a transparent PNG layer matching the first image’s dimensions and cannot be combined with a mask.
  • Output and planning: Supported formats are PNG, JPEG, WebP, animated GIF and SVG export for web, app and print work, per OrcaRouter and TestingCatalog. A three-level complexity parameter controls planning depth; OpenAI recommends low for quick edits, medium or high for dense scenes.
  • Pricing unchanged: OrcaRouter listed per-million-token rates identical to Images 2.5: text input at $5.00 (cached $1.25), image input at $8.00 (cached $2.00) and image output at $30.00. OpenAI’s pricing page lists both Flare and Sunburst.
  • Rollout and scale: The release is rolling out globally across ChatGPT and Codex on web, desktop and mobile at every subscription tier, though OrcaRouter said some accounts did not receive GIF on day one. Weekly output rose from over 3 billion images announced with Images 2.5.

Background:

Images 2.5 shipped on September 8, 2026, when OpenAI put weekly image generation above 3 billion across ChatGPT and the API, per unite.ai and OrcaRouter. The 2.6 release keeps the same token pricing while adding animation and vector export.

Between the lines:

The unchanged price list alongside two new models and SVG export points at distribution rather than margin: OpenAI is widening what the same spend buys. The two-model split also reads as a workflow bet — Flare for the fast layout pass, Sunburst for the finishing edits where brand detail has to survive repeated rounds. GIF output without GIF input keeps the system a generator, not an editor of existing animation.

What’s next

Watch whether Artificial Analysis publishes independent results for GPT-Image-2.6, which it had not listed as of November 5. Also track when GIF generation reaches accounts that missed it at launch.

Read on ontime+ ↗