ChatGPT Images 2.0

OpenAI has rolled out ChatGPT Images 2.0, a new image system that can reason through prompts and keep details consistent across multiple pictures. (openai.com) The company showed the update on April 21, 2026, and its developer docs now list `gpt-image-2` alongside earlier GPT Image models. OpenAI’s image-generation tool says the model can take text prompts, optional image inputs, and decide whether to generate a new image or edit one already in the conversation. (openai.com) (developers.openai.com) In plain terms, image reasoning means the model does more than paint pixels from a caption. It can use the same back-and-forth logic as a chatbot to decide what to change, what to preserve, and how separate images should match each other. (developers.openai.com 1) (developers.openai.com 2) That is the shift from the image tools OpenAI shipped in December 2025. That release, powered by GPT Image 1.5, focused on faster generation and more precise edits that kept lighting, composition, and people’s appearance intact across edits. (openai.com) The new API surface matters as much as the pictures. OpenAI’s docs say developers can use image generation through the Responses API for multi-step conversations or through the Image API for direct generation and edits, which puts image creation inside the same application flows already used for text and tools. (developers.openai.com 1) (developers.openai.com 2) That changes what software teams have to manage. When a model can read images, write images, revise prompts, and choose whether to edit or regenerate, platforms have to track which users get those features, how much each step costs, and what moderation rules apply at each stage. (developers.openai.com 1) (developers.openai.com 2) OpenAI’s current docs also show a split between what is newest in ChatGPT and what is fully documented for developers. The ChatGPT-facing model alias is `chatgpt-image-latest`, while the tool guide names `gpt-image-2`, and the broader image-generation guide still centers GPT Image 1.5 for the Image API. (developers.openai.com) (developers.openai.com) (developers.openai.com) OpenAI is also still winding down older image models. Its image-generation guide says DALL·E 2 and DALL·E 3 are deprecated and support will end on May 12, 2026, which puts more weight on the GPT Image line as the company’s default path for image work. (developers.openai.com) The practical result is that OpenAI is treating images less like a one-shot art feature and more like a native part of a multimodal assistant. The closer image generation moves to the core chat and API stack, the more it looks like standard infrastructure that developers will have to meter, govern, and watch in production. (developers.openai.com) (developers.openai.com)