OpenAI · Google · Image and video
OpenAI’s image model posts a near-perfect score on text-dense images
Researchers led by Westlake University released UltraText Bench on October 7. It asks image models to render dense English and Chinese text across 24 scene types, from newspapers and menus to invoices and code, with four to twelve text regions per prompt. Tested at its low quality setting, OpenAI’s GPT Image 2 ranked first among 24 configurations with 99.35 out of 100, while Google’s Nano Banana 2 came second with 97. The scores come from an AI judge model, and the authors note that a 100 is not a measured share of correct characters and that some top-rated images still show visible errors.
Sources
- arXiv, “UltraText Bench: A Comprehensive Bilingual Benchmark for Evaluating Visual Text Rendering in Image Generation”, (arxiv.org)
- OpenAI, “Introducing ChatGPT Images 2.0”, (openai.com)
- Google, “Nano Banana 2: Google’s latest AI image generation model”, (blog.google)
About this story
This story was posted on Instagram by @jarrus.tech on Oct. 10, 2026.
Spotted an error in this story? [email protected] · Instagram
Short link: thejarrus.com/en/gpt-image-2-text
This story in Turkish: OpenAI’ın görsel modeli, yoğun metin içeren görsellerde neredeyse kusursuz skor aldı
On the same topic
Robot tests: even the best model completed only 19% of 84 robot tasks, and 63 tasks were solved by no model
In RobotWorld, a benchmark released October 7, the top model, GPT-6 Astra, completed only 16 of 84 simulated robot tasks; no model solved 63 of the tasks.
A flaw in ChatGPT’s Mac app that could expose private chats was found and patched
Patrick Wardle found CVE-2026-100754, a flaw in ChatGPT’s Mac app that could expose chat history; OpenAI fixed it on September 25 in version 26.924.20706.
Higgsfield launches Ads Studio, turning a website into ad images for about 20 cents each
Higgsfield’s Ads Studio, announced October 1, builds a brand kit from a website and makes up to 20 static ads per product at about $0.20 each.