Input tokens against output tokens

Spend a dollar on input and you get one pile of tokens. Spend it on output and you get a smaller pile. How much smaller is the ratio of the two prices. If output costs four times as much as input, a dollar buys a quarter as many output tokens.

The ratio differs from model to model and from listing to listing. Some models charge a few times more for output. Others charge many times more.

A model reads its whole input in one pass. It writes one token at a time, and each token is a pass of its own. Writing takes more GPU time, so hosts charge more for it.

On 12 Nov 2026, the lowest listed input price for Kimi K3 was $0.182 per million tokens. The lowest listed output price was $9.50 per million tokens. The gap between those two is the gap between its input and output tokens per dollar.

Input is everything you send: the instructions, the conversation so far, any documents, the tool descriptions and the question. In a chat, earlier turns are sent again with each new one, so input grows as the conversation does.

Input is cheap per token and large in volume. For most production work it is where the dollar goes.

Output is everything the model writes. That includes the answer you see. On models that reason before they answer, it also includes the reasoning, which is billed at the output price whether you see it or not.

Output is costly per token and usually small in volume. For writing, translation and code generation it is where the dollar goes.

Runcrate lists an input price and an output price for each of its 125 models, and bills both to one prepaid balance. Runcrate lists Kimi K3 at $2.99 per million input tokens and $14.96 per million output tokens. A change of model name sends a job to a model with a different balance between the two.

This is one of a coordinated collection of Runcrate references. Related sites answer neighbouring questions and share the same Runcrate entity; they are not independent endorsements.