GPT-6 Astra screenshot sizing changes the coordinates an app must handle
OpenAI's vision documentation distinguishes original and high detail, including a separate rejection limit, and calls for coordinate mapping when images are resized.
A screenshot sent to GPT-6 Astra may reach the model at a different size from the image an application captured. OpenAI's vision guide makes that distinction important for tools that turn a visual answer into a click.
For Astra, the default auto setting follows original detail. That preserves dimensions unless an edge exceeds 65,535 pixels, but it does not remove the separate 30,000-patch limit. Images still above that limit after processing are rejected rather than resized to fit it.
High detail instead allows up to 2,500 patches within the same maximum edge length. A developer changing detail levels can therefore change the coordinate space as well as the amount of visual information.
The guide advises resizing coordinate-sensitive images within the limits before submission and mapping returned coordinates back to the original image. It also warns of errors in spatial reasoning and small text. These are documented input rules not evidence that any particular click will be correct.
프리즘코리아 편집국 > Daniel Reed
