You don't even need to do that- pay a subscription, I mean. A gemma 3 4b model will run on near potato hardware at usable speeds and achieves performance for many purposes on part with ChatGPT 3.5 turbo or better in many tasks much more beneficial than ad tech and min/max'ing media engagement. Or the free versions of many SOTA web LLMs, all free, to the world, if you have a web browser.