Model provider

Running a frontier model takes a rack of expensive accelerators and a lot of engineering. Almost nobody does it themselves. Instead you use a model provider: a company like Anthropic, OpenAI, or Google that hosts the model and exposes it over an API. Your coding tool sends text to the provider, the provider runs inference on its own hardware, and the result comes back over the wire.

What the provider actually owns

Understanding the split helps you reason about cost, privacy, and reliability:

• The weights and the hardware. The model file and the machines it runs on live with the provider, not on your laptop.
• The API contract. The provider defines the request shape, the available models, the rate limits, and the pricing.
• Uptime and versions. When the provider has an outage or retires a model version, your tool is affected, because your tool is just a client.

Why it matters when you are coding

Your harness is a thin layer sitting on top of somebody else's API. That is freeing and constraining at once. You get state-of-the-art models without owning any infrastructure, but you also inherit their latency, their pricing, and their limits. Every request leaves your machine and is processed on the provider's servers, which is worth remembering before you paste in secrets or proprietary code.