OpenAI introduces Images 2.0 with “thinking” capabilities for image generation

OpenAI has introduced Images 2.0, its first image generation model with “thinking” capabilities, according to an official announcement.

When a reasoning model is selected in ChatGPT, Images 2.0 can search for relevant information online, generate multiple images from a single prompt, and verify its own outputs.

Compared with earlier models, Images 2.0 shows stronger performance in non-Latin scripts, including Japanese, Korean, Chinese, Hindi and Bengali. Text in these languages is not only rendered correctly but also appears coherent and contextually accurate.

The model also improves visual fidelity. It can introduce subtle imperfections to make images appear more realistic and more precisely reproduce cinematic frames, pixel art and other visual styles.

OpenAI notes that the model may still struggle with certain tasks, including generating step-by-step instructions such as origami, puzzles, or rendering fine details on tilted surfaces.

Images 2.0 is available to all ChatGPT and Codex users. The reasoning mode is limited to ChatGPT Plus, Pro and Business subscribers. The gpt-image-2 model is also available via API.