OpenAI on Tuesday released ChatGPT Images 2.0, a new image generation model it describes as a "step change" over its predecessor. The update adds reasoning capabilities to an image model for the first time, allowing the system to search the web, compile information, and check its own outputs for accuracy.
The model is designed around practical, text-heavy tasks: infographics, scientific posters, study guides, and marketing materials. That focus reflects a deliberate shift from OpenAI's earlier generative media experiments. The company shut down its Sora AI video app last month to concentrate on what it calls "core products," and Images 2.0 fits that direction.
One of the more notable improvements is how the model handles non-Latin scripts. OpenAI says it made significant gains with Japanese, Korean, Chinese, Hindi, and Bengali, areas where earlier image models routinely failed. The company also says the model better reproduces the visual characteristics of different design traditions, which it argues makes the tool more useful for game prototyping and storyboarding.
On the technical side, Images 2.0 supports aspect ratios as wide as 3:1 and as tall as 1:3. Developers using the API can generate images at 2K and 4K resolution, though the higher resolution option remains in beta. The model can produce up to eight image outputs in a single request.
Engadget got early access to the model and tested it with a pixel art prompt based on the Game Boy Advance Pokémon art style. The reviewer called the result "commendable" and noted the model successfully generated a transparent PNG, something other image models can struggle with. A four-panel manga was also generated as part of the preview.
The model is available now to all ChatGPT users, including those on the free tier. Paid subscribers on Plus and Pro plans receive higher generation limits and access to the reasoning-enhanced outputs. Adele Li, product lead for ChatGPT Images, described the broader vision as building a "creative assistant" within ChatGPT's goal of becoming an all-purpose personal AI tool.
