ChatGPT’s new Images 2.0 model is surprisingly good at generating text

OpenAI has released ChatGPT Images 2.0, a major upgrade that finally solves one of AI image generation's most persistent flaws: rendering legible text. Where previous models like DALL-E 3 would produce garbled nonsense like "burrto" or "margartas" on a restaurant menu, Images 2.0 creates photorealistic, usable marketing materials. The breakthrough stems from a new architecture—likely an autoregressive model that functions more like an LLM—which gives it "thinking capabilities" to search, reason, and double-check its outputs for accuracy.

This new approach allows the model to handle complex, multi-step tasks. It can generate a series of images from a single prompt, create marketing assets in various sizes, and produce detailed multi-panel comic strips. Crucially, it has a stronger understanding of non-Latin scripts, accurately rendering text in Japanese, Korean, Hindi, and Bengali. Outputs can reach up to 2K resolution, preserving fine details like small text and UI elements that typically break other models.

The model's knowledge is current through December 2025, and while complex generations take a few minutes, the fidelity is unprecedented. All ChatGPT users gain access, with advanced outputs reserved for paid tiers. OpenAI is also releasing a gpt-image-2 API, with pricing based on output quality and resolution. This shift turns AI image generation from a novelty into a practical tool for professional design and content creation.