Overview
- ChatGPT Images 2.0, unveiled Tuesday, is now live across ChatGPT, Codex, and the gpt-image-2 API with basic access for all users and advanced features for Plus, Pro, Business, and Enterprise tiers.
- When paired with a thinking model, the system can search the web, plan an image before it renders, verify outputs, and generate up to eight related images that keep characters and styles consistent.
- The model supports outputs up to 2K resolution and flexible aspect ratios from 3:1 to 1:3, aiding assets like banners, slides, posters, and mobile graphics without extra editing.
- OpenAI says it fixed long‑standing pain points such as warped or misspelled text and messy layouts, adding stronger multilingual rendering for non‑Latin scripts like Japanese, Korean, Chinese, Hindi, and Bengali; partner platforms such as Pollo AI have begun integrating the model.
- The company notes limits on tasks that need precise physical reasoning, on very dense or repetitive textures, and on some detailed diagrams; the model carries a December 2025 knowledge cutoff even as web search can supply live context when thinking is enabled.