OpenAI laid out a set of numbers in its official blog: today, users worldwide create 4 billion images per week through ChatGPT, Codex, and the API, averaging more than 500 million per day.
When AI image generation reaches this scale, the industry's pain point is no longer "can it draw it," but "once it's drawn, can it be easily edited?" Following this shift, OpenAI today officially launched Images2.6 and simultaneously made it available in ChatGPT, Codex, and the API, covering all subscription tiers and desktop, mobile, and web.
Faster, animated, and more consistent to edit
This upgrade includes three main changes. First, generation speed is increased by about 50%, allowing users to revise prompts more frequently and fine-tune images repeatedly. Second, it officially supports generating GIF animations from text prompts. The model can understand continuous motion logic and orchestrate character settings, fixed camera, frame structure, and loop transitions together, taking a step from static images toward lightweight animation.
Third, it strengthens consistency across multiple rounds of editing—the most grueling part of real-world creation, where a single edit often distorts a character's face or a product's appearance. For developers, OpenAI offers two API models with different positioning: Flare, which prioritizes speed, can generate an image in as little as 6 seconds at low quality; Sunburst, which emphasizes image quality and consistency, is responsible for preserving character, product, and brand details through successive edits. This is especially critical for teams creating series posters and brand assets.
Giving control back to the workflow
The more substantial updates are hidden in the editing stage. Images2.6 brings the image marking feature already available in ChatGPT to the API, allowing developers to precisely specify "which part to change" and avoid a small fix turning into a full-image redraw; it also supports SVG export, making it easy for results to go directly into websites, apps, and print scenarios. The newly added complexity parameter lets developers control how much planning the model invests to satisfy a prompt—use a low value for simple background swaps in exchange for speed, and medium-to-high values for complex compositions in exchange for precision.
Pricing remains the same as the previous Images2.5: $8 per million tokens for image input and $30 for output. Separately, according to The Information, citing people familiar with the matter, OpenAI could launch Astra, Sol, and a cybersecurity model as early as Friday. If true, Images2.6 may be just the opening act in a string of updates this week.
