API
Everything you build on top of AI runs through this — understanding it makes you faster
What is an API
Every time you check the weather on your phone, your phone did not go outside.
It sent a request to a server somewhere — a computer that tracks weather data. The server sent back the current temperature, the forecast, the wind speed. Your phone showed it to you. You never saw the request or the response. You just saw the weather.
That exchange — one system asking another for data, getting a structured answer back — is an API. Application Programming Interface. The interface part is the contract: ask in this format, receive an answer in that format. Every time, reliably, at scale.
Why developers live inside APIs
When you use Claude 3.5 Sonnet in a product you build, you are not running a model on your laptop. You send a message to Anthropic's servers through their API: here is my prompt, here are my settings, send me a response. They process it and send one back. You show it to your user.
The model stays on Anthropic's infrastructure. You use the interface.
Everything on sourc.dev — input prices, context windows, capability flags — describes the API for each model. The price is what the API charges. The context window is how much the API accepts per call. The rate limit is how many times per minute you can call it.
Why this matters to you
APIs are why a two-person team can build a product powered by technology that cost hundreds of millions to train. You did not build the model. You did not buy the servers. You called the API.
Understanding what an API is — and is not — shapes every architectural decision you make when building with AI. It is not software you install. It is not a service you log into. It is a set of rules for how two systems communicate, and a remote server that follows those rules on demand.
One codebase. One API call pattern. Scales to ten users or ten million. The infrastructure on the other side handles it. Yours does not have to.
Verified March 2026