OpenAI's ChatGPT Images 2.0 Leaks on LM Arena AI: Multilingual Text, Full Infographics, and Flawless UI Generation

OpenAI has been secretly testing a dramatic new AI image model, ChatGPT Images 2.0, for weeks under the codename "duct tape" on the third-party platform LM Arena AI. This update, following the GPT-Image-1.5 release from December 2025, represents a significant leap in capability, already astonishing early testers with its ability to generate complex, multi-element images that previous models struggled with.

The new model's most striking features include the flawless generation of long, coherent blocks of multilingual text and disparate text panels within a single image. It also demonstrates an uncanny ability to produce insanely realistic user interfaces, screenshots of popular websites, and detailed reproductions of real-life scenes. This suggests a major advancement in instruction following and compositional understanding, moving far beyond simple object generation to create complete, functional visual documents like infographics, slides, maps, and even manga-style panels.

This pre-release testing on LM Arena AI, a platform used by major AI providers for early feedback, signals OpenAI's aggressive push to dominate the multimodal AI space. The capacity to generate such complex, text-heavy graphics flawlessly could disrupt industries reliant on design and content creation, from marketing and publishing to software prototyping. The leak of its capabilities also intensifies the competitive pressure on other AI labs, setting a new public benchmark for what is considered state-of-the-art in generative imagery.