Input token

Tokens you send to the model. Usually 3-10x cheaper than output.

Input tokens are everything you send to the model: system prompt, tool definitions, conversation history, current user message, and any attached images (converted to tokens). Input is almost always cheaper per token than output, typically by a factor of 3-10x. Cached input tokens can be another 4-10x cheaper depending on provider. Because input dominates the token count on repetitive workloads (chat histories, repo reads), the input price and cache discount usually matter more for total cost than the headline output price.