Input Tokens
The tokens in everything you send the model — the prompt, system message, tools, and prior turns — billed at the input rate.
Input tokens are every token the model reads before it starts generating: your prompt, the system message, any tool or function definitions, retrieved documents, and the full conversation history you resend on each turn. Providers meter them separately from output and almost always price them lower, because reading text is cheaper than generating it.
On this explorer, input is the first price column and is quoted in US dollars per one million tokens. It is usually the number that matters most for retrieval-augmented and long-context workloads, where the prompt dwarfs the answer — a 50-page document you pass in on every call is billed as input every time unless you use prompt caching.
Because input dominates so many real workloads, a model with a low input rate but a high output rate can still be the cheaper choice for summarization or extraction, and the reverse holds for chat or code generation. Use the token-cost calculator on the hub to weigh both against your actual input/output mix rather than trusting a single sticker number.
FAQ
Are input tokens cheaper than output tokens?
Almost always. Every provider in this explorer prices input below output — often three to six times lower — because processing a prompt costs less than generating new text.
Does the system prompt count as input?
Yes. The system message, tool definitions, and the entire prior conversation you resend are all billed as input tokens on each request.