OpenAI Launches ChatGPT Images 2.0

On April 21, 2026, OpenAI released the next generation image generation model, ChatGPT Images 2.0. This model integrates reasoning capabilities into image generation, achieving a score of 2240 points on the Arena.AI image generation rankings, surpassing the second-place competitor, Nano Banana.

According to the official introduction, ChatGPT Images 2.0 is OpenAI’s first image model with “thinking” capabilities. Once enabled, the system can plan the structure of images through reasoning before generation and automatically retrieve information online to complete details such as brands and scenes. For multi-image generation, users can output up to eight images in a single prompt while maintaining consistency in characters, objects, and styles. The new version supports a maximum resolution of 2K and expands the aspect ratio range to 3:1 and 1:3, optimizing rendering accuracy for non-Latin scripts like Chinese, Japanese, and Korean.

OpenAI positions this upgrade as a transition from a rendering tool to a visual system. Currently, ChatGPT produces over 1 billion images weekly, and this new model further targets the professional user market, capable of tasks such as product advertisement design, academic poster generation, and UI interface creation. Official demonstrations show that simple prompts can generate highly realistic interface screenshots and social media graphics. All ChatGPT and Codex users can now access the basic version for free, while advanced outputs with reasoning capabilities are available to Plus, Pro, Business, and Enterprise subscribers. The API charges based on quality and resolution. Due to the multi-step reasoning involved in the Pro version, operational costs are significantly higher, making the standard version more economical for everyday lightweight tasks.

The model’s knowledge cutoff is December 2025, and the development team is led by Gabriel Goh, including Chinese researchers such as OpenAI research scientist Chen Boyuan.