The AI model o3 has demonstrated advanced capabilities in image analysis and reasoning through various tasks. It successfully identified individual plants in a flower garden photo, correctly naming 10 out of 15 species. In another instance, o3 solved a crossword puzzle by analyzing the image multiple times and using external web searches, completing the task in 11 minutes and 6 seconds with only a minor positional error. Additionally, o3 decoded a puzzle involving Nvidia's H100/A100 flop ratio by cropping and researching relevant data online. The model also identified and located a Chinese restaurant menu in San Francisco without any metadata, relying solely on web searches to match menu items. These examples highlight o3’s ability to use tool-assisted image zooming, cropping, and web exploration to perform complex reasoning tasks that resemble human-like attention to detail.