Back to glossary

Term

Input Token

Tokens you send to an AI model in an API call — your prompt, the context, attached documents. Billed separately from output tokens and usually much cheaper.

Input token — in more detail

Input tokens are everything you hand the model on a call: system prompt, user question, attached context (documents, code, conversation history), tool definitions. Before the model can respond, it has to read these tokens and convert them into its internal representations. That step is what is billed as input tokens — typically between $0.15 and $3 per million tokens, depending on model and provider.

Example / practical context

A typical coding request with 30,000 tokens of repository context plus a short user question consists almost entirely of input tokens. With Claude Sonnet ($3/MTok input), that single request already costs $0.09 in input — multiplied by hundreds of requests per day, it becomes the dominant cost line. Whoever reduces input tokens (smaller contexts, targeted chunking, prompt caching) saves immediately.

Output tokens are what the model sends back — and are usually three to five times more expensive than input tokens. Cached input tokens are billed at a steep discount when the same prompt prefix is sent repeatedly (often just 10 % of the regular input price). The sum of input and output tokens must stay within the model’s context window.

Entdecke mehr