--- title: "What's New in OpenAI Images 2.6? GIFs, Edits and Speed Tested" url: "https://goremy.ai/blog/articles/openai-images-2-6-animated-gifs-speed" author: Luis Chavez-Mattos editor: Luis Chavez-Mattos editor_title: Director, Product published: "2026-11-10T08:20:09.529Z" updated: "2026-11-10T17:02:19.892Z" topic: AI Image, Video & Audio description: "OpenAI's Images 2.6 in ChatGPT and the API makes animated GIFs, handles follow-up edits, and generated a standard image in about 20 seconds in one test." --- # What's New in OpenAI Images 2.6? GIFs, Edits and Speed Tested > OpenAI's Images 2.6 in ChatGPT and the API makes animated GIFs, handles follow-up edits, and generated a standard image in about 20 seconds in one test. Edited by Luis Chavez-Mattos, Director, Product ยท Published 2026-11-10 ## What is new in OpenAI Images 2.6? Images 2.6 is OpenAI's latest image model, available in both ChatGPT and the API. OpenAI pitches it as faster and better at professional-looking visuals. The headline feature is animated GIF generation. In one hands-on test, a GIF prompt returned in about 20 seconds, follow-up edits took around 30 seconds, and a standard still image took 20.4 seconds. ## TL;DR - Images 2.6 is available in **both ChatGPT and the API**, so you can use it in the app or build on it. - The new trick is **animated GIF generation** from a plain text prompt. - In a ChatGPT test, a roller-skating monkey GIF came back in **roughly 20 seconds**. - **Follow-up edits work on GIFs.** Adding a baseball hat and moving the scene to a street took about 30 seconds. - A regular still image (a San Diego skyline at sunset) finished in **20.4 seconds**, which felt faster than the previous model. - The tester called it **a clear speed improvement** over the previous model, but no older-model timing was given and this was a single informal test. ## How does GIF generation work in ChatGPT? You type a prompt that asks for an animated GIF. The tester used "make an animated GIF of a monkey roller skating." About 20 seconds later, ChatGPT returned a looping animation of a monkey on roller skates. Nothing special is needed beyond saying "animated GIF" in the request. The result appears in the chat like any other generated image. Clicking it opens an option to edit. OpenAI has not published, in the material reviewed here, how the GIF is built. The tester guessed that the model makes several frames and stitches them together. That is speculation, but it fits how short looping animations usually work. Treat it as a mental model, not a documented fact. ## Can you edit an animated GIF after it is made? Yes. Clicking the generated GIF opens an edit option, and follow-up prompts apply to the animation. The first output placed the scene in San Diego. The tester suspected this came from ChatGPT's memory, since they live there. That is worth knowing: memory and personalization can shape image output even when your prompt never mentions a place. Two edit prompts went in back to back: 1. Make the monkey wear a baseball hat. 2. Make him roller skate in the street. About 30 seconds later, the monkey had a hat and was skating on a street instead of a sidewalk. The edit held the same character and concept while changing the requested details. For anyone iterating on a visual, that matters more than raw speed. You can steer a result instead of rerolling from scratch. ## How fast is Images 2.6? Two timings came out of the test. | Task | Approximate time | |---|---| | Initial animated GIF (roller-skating monkey) | ~20 seconds | | Two follow-up edits on the GIF | ~30 seconds or less | | Still image (San Diego skyline, sunset, sailboats) | 20.4 seconds | The still image was timed with a manual stopwatch from submit to finish. It looked solid and, by the tester's account, was a clear speed improvement over the previous OpenAI image model. No exact figure for the older model was given, so the size of the gain is not measured here. One oddity: the GIF arrived in about the same time as a single still. If GIFs really are built from multiple frames, that is a notable result. Still, one prompt on one account at one time of day tells you little about typical latency. Server load, prompt complexity and plan tier can all shift it. ## Is the image quality good? For the still image, the tester called it "pretty solid." The skyline prompt produced a recognizable San Diego scene with sailboats on the water. For comparison, the same prompt was run on a free open-weight image model (Pruna AI's PH3 image, built from a video model) in a Hugging Face Space. The result looked nothing like the San Diego skyline. The tester said its quality was nowhere near ChatGPT's. That is one prompt, so it is not a benchmark. It does show that a famous real-world location is a useful stress test for any image model. Open-weight models offer other things. You can download them, fine-tune them and run them locally. PH3 image weights are around 55 GB for the full version, with smaller quantized versions available. The tradeoff against a hosted model like Images 2.6 is control versus out-of-the-box quality. ## Is Images 2.6 worth trying? It depends on what you make. **Worth testing if you:** - Need quick looping animations for social posts, docs or prototypes without opening animation software. - Iterate on visuals through chat and want edits that preserve the subject. - Build products on the API and want GIF output as a feature. **Probably wait if you:** - Need precise frame control, long animations or video. GIFs are short loops, and dedicated video models are a different category. - Need hard numbers on API pricing or rate limits. The source material here does not give them, so check OpenAI's documentation before planning around it. ## What should builders check before using it? The API availability is the part developers care about, but this review did not cover API details. Before building, confirm a few things in OpenAI's docs: - Whether GIF output is exposed through the API the same way as in ChatGPT. - Pricing per image and any difference for animated output. - File size and resolution limits for GIFs. - How editing is handled programmatically, since the ChatGPT flow is conversational. Run your own timing tests with your own prompts. A 20-second result is a useful baseline, not a guarantee. ## Frequently Asked Questions ### Where can I use OpenAI Images 2.6? It is available in ChatGPT and in the API. In ChatGPT you can request an image or animated GIF directly in a chat and then use the edit option on the result. ### How long does Images 2.6 take to make a GIF? In one test, a prompt for a roller-skating monkey returned a GIF in about 20 seconds. Two follow-up edits took roughly 30 seconds combined. Your times may vary with load and prompt. ### Does Images 2.6 edit existing GIFs? Yes. The tester clicked a generated GIF, chose to edit, and asked for a baseball hat and a street setting. Both changes applied while the core scene stayed intact. ### Is Images 2.6 really twice as fast? The article gives no 'twice as fast' figure. The hands-on test found a 20.4-second still image and called it a speed improvement, but it did not measure the older model directly, so the size of the gain is unknown.