That's basically model routing, and has existed a while. Open AI's GPT-5 and llama-swap do that, for example. If the task is simple, it uses a smaller, less intensive model, and only uses the slower, larger one of the task is more complex.
Though most tend to operate with models on the same device/service, rather than a model run elsewhere.
We had a rather nice thing going with pure HTML. Sure, it wasn't the prettiest thing, even with CSS, but almost every device could run and display it in its own way.
You didn't need a custom thing, or a bunch of extra code adjusting the webpage for each type of device that opened the web page, since that job was all done by the browser.