Skip to main content

They are not the same model, and the text-only model is definitively able to output graphical content via various forms of code; see e.g. Sparks of Artificial General Intelligence: Early experiments with GPT-4 for a source on this and for some concrete examples.