Home / Docs / Glossary

Glossary

The terms used throughout these docs, defined precisely.

TermMeaning
TokenThe unit models read and write, roughly ¾ of an English word. Billing is per million tokens.
Input tokensEverything you send: system prompt, full message history, tool schemas, images. Grows every turn of a conversation.
Output tokensWhat the model generates. Costs 4–5× input on every frontier model.
Context windowThe maximum tokens a model can consider at once, input plus output. Exceeding it is an error, not silent truncation.
Base URLThe origin a client sends requests to. The single most common misconfiguration — see Base URLs.
Messages APIAnthropic's request format. POST /v1/messages, replies in a content[] array.
Chat CompletionsOpenAI's request format. POST /v1/chat/completions, replies in choices[].
Streaming / SSEServer-sent events delivering the reply token by token instead of in one body.
stop_reasonWhy generation ended: end_turn, max_tokens, stop_sequence or tool_use.
Prompt cachingRe-reading a stable prefix at a reduced input rate. See Prompt caching.
Tool useThe model requesting that you run a function, then continuing with the result.
MCPModel Context Protocol — a standard way to expose tools to an agent.
CreditsPrepaid balance spent at per-token rates. Do not expire.
Unlimited planFlat-rate access for a fixed window, under fair-use limits, instead of per-token billing.
GatewayA service that fronts one or more model providers behind a single API and key.