The arrival of ChatGPT Images 2.0 challenges the very trend that Google’s Nano Banana introduced last year. The update finally promises to solve one of the longest-standing frustrations in AI-generated visuals, i.e., creating clean, accurate text inside images. The improvements that OpenAI has made with the new Images 2.0 update promise to address the basic text rendering issues, with capabilities that are superior to Google’s Nano Banana 2 and Nano Banana Pro.
Users are already showing off the image generation prowess of the updated ChatGPT. One user managed to generate complex restaurant menu cards with properly legible texts, rendered with the matching art style. But there’s more to Images 2.0 than simply printing menu cards.
Here’s a fun fact: the cover image you see in this article has been generated by ChatGPT Images 2.0. Check out the text and its legibility.
What is ChatGPT Images 2.0 doing better than before
ChatGPT Images 2.0 (powered by OpenAI’s gpt-image-2 model) currently leads major blind-voting leaderboards such as LM Arena with an Elo score of 1,512 — a record 242-point jump, which is far ahead of Google’s Nano Banana 2 (Gemini 3.1 Flash Image Preview) at 1,270. On Artificial Analysis, the prior GPT Image 1.5 scores approximately 1,273 Elo, narrowly ahead of Nano Banana 2 at 1,264.
ChatGPT Images 2.0 outperforms Nano Banana 2, Ideogram v3, Midjourney, and Flux.2 variants in dense text rendering, complex typography, multilingual support (including non-Latin scripts like Hindi, Japanese, and Korean), and precise instruction-following for UI elements, icons, and multi-panel layouts. While Nano Banana 2 leads in generation speed (3-20s per image) and photorealistic portraits, ChatGPT Images 2.0 delivers superior text accuracy, publication-ready output, and consistency across image batches.
