There are two worlds here, and people muddle them constantly. Getting them straight saves you a lot of confusion later.
Frontier cloud (rented). OpenAI, Google, Anthropic, Microsoft and the rest — enormous models running on their servers. You pay a subscription or per-use API fees. You get the best polish and the newest features first — under their rules.
Open-weight (ownable). Llama, Mistral, Qwen and others — you download the brain file and run it on your own machine or server. Your hardware, your data policy — and a bit more elbow grease required.
Neither one is magic. The rented route is account + app, their servers, £/month plus caps. The owned route is install + download, your machine, your electricity bill. The very top capability is often still frontier cloud — but open-weight is catching up fast, and which one wins genuinely depends on the task.
Here's how that maps onto this site: later in Ground Zero you'll compare the rented apps hands-on. Off the Grid installs Ollama on your own PC. And membership builds your Jarvis — deliberately mixing local models with a bit of pay-as-you-go cloud, using each where it's strongest.
Continue — the speed of it all.