OpenAI has released ChatGPT Images 2.0, a new version of its AI image generation tool available to all ChatGPT and Codex users. The model, powered by a new underlying system called GPT Image 2, introduces what the company calls “thinking capabilities,” allowing it to search the web before generating images. Paid subscribers on Plus, Pro, Business, and Enterprise plans get access to a more advanced version of these features.
The most significant change in Images 2.0 is its ability to generate multiple images from a single prompt — up to eight at once — while keeping characters, objects, and styles consistent across each image. OpenAI says this makes it practical to produce things like comic strips, manga pages, series of social media graphics, or design plans for multiple rooms in a house.
Better text, higher resolution, more aspect ratios
One area where earlier AI image models consistently fell short was the rendering of readable text. TechCrunch noted that just two years ago, asking an AI image generator to produce a restaurant menu resulted in invented words like “enchuita” and “burrto.” In testing, ChatGPT Images 2.0 produced a legible, realistic-looking Mexican food menu without such errors.
Wired tested the model and found text rendering in English to be noticeably improved. Testing in other languages produced mixed results. When Wired asked the model to generate a Chinese-language fan poster, ChatGPT itself acknowledged that much of the text it had produced was “fake, or semi-gibberish AI text dressed up to look like Chinese meme-poster writing.”
OpenAI says the model makes significant improvements in non-Latin scripts, including Japanese, Korean, Chinese, Hindi, and Bengali, though independent testing suggests these gains are uneven.
Other technical improvements include a maximum resolution of 2K and support for a wider range of aspect ratios, from 3:1 wide to 1:3 tall. The model’s knowledge cuts off in December 2025, meaning prompts that rely on very recent events may not produce accurate results.
Thinking mode adds web research before image creation
When the thinking mode is enabled, the model can pull live information from the web and use it to inform image generation. In testing by Wired, a prompt asking for a weather forecast infographic for San Francisco returned an image with accurate weather data and recognizable local landmarks. The model can also analyze files a user uploads and use that content to generate visual explainers.
OpenAI describes Images 2.0 as capable of handling “dense compositions,” fine-grained details, UI elements, and small text within images. These elements have traditionally caused problems for image generation models. The company says generation of complex outputs, such as multi-panel comics, takes a few minutes.
Availability and pricing
ChatGPT Images 2.0 is available to all ChatGPT and Codex users. Paid subscribers have access to higher-quality outputs and the thinking capabilities. OpenAI has also made the underlying model available through its API under the name gpt-image-2, with pricing based on the quality and resolution of outputs.
Prompt for the infographic:
Research the latest news and reactions to “ChatGPT Image 2.0”, introduced today, April 21 2026, by OpenAI. Design a modern and stylish infographic in 4:5 portrait aspect ratio that gives an overview of its main capabilities and what others have said and written about it. My brand’s color is D8405C and please use this as the main color for accents.
Sources: OpenAI, TechCrunch, The Verge, 9to5Mac, Wired
Stay up to date
AI for content creation: the latest tools, tips and trends. Every two weeks in your inbox:
