OpenAI Launches ChatGPT Images 2.0 With Reasoning Capabilities and Sharply Better Text
The new model generates accurate multilingual text, high-fidelity marketing assets, and multi-panel comics — and is available to all ChatGPT users starting Tuesday.
OpenAI released ChatGPT Images 2.0 on April 21, describing it as a generational advance over its previous image-generation models. Unlike the diffusion-based DALL-E 3, the new model incorporates reasoning capabilities, allowing it to perform web searches during generation, check its own outputs, and produce multiple images from a single prompt.
The most striking improvement is text rendering. Where earlier models routinely hallucinated garbled words — producing images of menus filled with misspelled dishes or nonsensical signage — Images 2.0 can accurately reproduce dense, legible text in English and a range of non-Latin scripts including Japanese, Korean, Hindi, and Bengali.
The model can generate realistic restaurant menus, UI mockups, icon sets, and dense compositions with small typography.
The model supports resolutions up to 2K and can produce marketing assets in a variety of standard sizes. Complex outputs such as multi-paneled comic strips may take a few minutes to render.
OpenAI has also made the underlying gpt-image-2 API available to developers, with pricing based on output quality and resolution.
All ChatGPT and Codex users gained access starting Tuesday. Paid subscribers on Plus, Pro, Business, and Enterprise plans receive access to higher-quality outputs.
OpenAI noted that the model carries a knowledge cutoff of December 2025, which may affect prompts referencing very recent events. The launch positions OpenAI more directly against dedicated image-generation competitors as the company continues consolidating its suite of creative tools within ChatGPT.
Read the original reporting at TechCrunch.
