How AI token pricing works, and what "blended price" means

API models are billed per token, with separate prices for the tokens you send (input) and the tokens the model writes (output). Prices are quoted per million tokens.

Why output costs more

Generating a token takes much more compute than reading one, so output is typically 3–5× more expensive than input. A model listed at $2 / $10 charges $2 per million input tokens and $10 per million output tokens.

Blended price

To compare models with one number, ModelBench uses a blended price that assumes three input tokens for every output token:

blended = (3 × input price + 1 × output price) ÷ 4

So $2 / $10 blends to (6 + 10) ÷ 4 = $4.00 per million tokens. The 3:1 mix is typical of chat and retrieval workloads, but yours may differ.

Estimate your own bill

The pricing calculator takes your monthly input and output volume and prices every model at once. Some rules of thumb:

• Retrieval (RAG) and long documents are input-heavy: the input price dominates.
• Code generation and long-form writing are output-heavy: look at the output price.
• Reasoning models produce extra "thinking" tokens that are billed as output, so their real cost per task can be several times their list price suggests.

What list prices leave out

Prompt caching, batch discounts and volume deals can cut costs substantially. The calculator uses list prices only, so treat it as an upper-bound starting point.