ChatGPT Images 2.0 debuts
- OpenAI said April 21 it launched ChatGPT Images 2.0 in ChatGPT, adding a new image model with better text rendering, multilingual output, and visual reasoning. - OpenAI also released gpt-image-2 for developers in the API and Codex, with up to 2K outputs, more aspect ratios, and a reported No. 1 Image Arena ranking. - The update expands OpenAI’s push beyond GPT Image 1.5, while adding new safeguards for more realistic synthetic images. (openai.com)
OpenAI launched ChatGPT Images 2.0 on April 21, adding a new image model inside ChatGPT with better text rendering, multilingual output, and visual reasoning. (openai.com) Image generation systems turn text prompts into pictures, and newer versions try to do more than make pretty scenes. OpenAI said this model is built to produce readable text, follow detailed instructions, and handle more complex layouts. (openai.com) (community.openai.com) The release also adds “images with thinking” in ChatGPT, which lets the system spend more time planning and refining an image before it generates one. OpenAI said that feature is available on paid plans when users choose Thinking or Pro models. (help.openai.com) (deploymentsafety.openai.com) For developers, the same launch came with gpt-image-2 in the application programming interface and in Codex. OpenAI said the model is aimed at production work, including ads, product flows, presentations, documents, and other formatted visuals. (community.openai.com) OpenAI said gpt-image-2 supports more aspect ratios, resolutions up to 2K, stronger editing, and better multilingual text rendering. The company said the goal is to make images usable without heavy cleanup after generation. (community.openai.com) The company is positioning this as the next step after its December 16, 2025 rollout of GPT Image 1.5 in ChatGPT. That earlier release emphasized faster generation and more precise edits that preserved details across revisions. (openai.com) OpenAI’s safety documents say the newer model’s added realism raises the risk of more convincing fake images of real people, places, and events. The company said it uses prompt filters, image-input checks, and output screening to block policy-violating results. (deploymentsafety.openai.com) OpenAI also said the new thinking mode can combine reasoning, tool use, and live web search data while developing an image. In the system card, the company described that workflow as a way to turn a basic prompt into a more researched final picture. (deploymentsafety.openai.com) The rollout is broad inside ChatGPT. OpenAI’s release notes said ChatGPT Images 2.0 is available on all ChatGPT plans, while the more deliberate “images with thinking” mode is limited to paid users. (help.openai.com) OpenAI said independent evaluations put gpt-image-2 at No. 1 across Image Arena leaderboards just after launch, including a 242-point lead in text-to-image. The release gives OpenAI a new flagship image model in both consumer ChatGPT and developer tools at the same time. (community.openai.com)