Token cost calculator
Prices are USD per 1 million tokens. Cache pricing applies to the share of input tokens served from cache, where supported.
Model price comparison
Click a column header to sort. Click a model row to see notes. Last verified: October 2026
| Provider | Model | Input $/1M | Output $/1M | Cached input $/1M | Context |
|---|
Prices are public list prices per 1M tokens, standard tier, and change often. Verify with the provider before budgeting. Cached input is shown where the provider publishes it; a dash means no published cached rate was found.
Token converter
Rule of thumb: 1 token ≈ 0.75 words in English; 1 page ≈ 500 words ≈ 667 tokens. Token counts vary by language and tokenizer.
Guides
How token pricing actually works
Every serious LLM provider bills per token, split into input tokens (your prompt plus context) and output tokens (the model's reply). Output almost always costs more than input — typically 4 to 6 times as much — because generating text is the expensive part. Many providers also charge less for cached input: when a repeated prompt prefix is reused, you pay a fraction of the normal input rate. Some offer batch APIs at around 50% off in exchange for slower turnaround. The number that matters for budgeting is your blended rate: the weighted mix of input, cached input and output your workload really produces, not the headline input price.
Five ways to cut your token bill
First, turn on prompt caching for repeated system prompts and long documents — cache reads can cost a tenth of fresh input. Second, use batch APIs for anything that is not latency-sensitive; the discount is often 50%. Third, route simple work to small models and reserve flagships for hard reasoning — a cheap model classifying or extracting text costs a fraction of a flagship doing the same job. Fourth, compress prompts: cut boilerplate, summarize long context before sending it, and stop re-sending the same background on every call. Fifth, cap output length where a short answer will do, since output tokens dominate most bills.
Why output tokens dominate your bill
A coding or reasoning assistant writes long replies, so output volume is high and output rates are the highest line on the price sheet. A request with 10,000 input tokens and 2,000 output tokens on a $2/$10 model costs $0.02 in and $0.02 out — already half the bill from one-sixth of the tokens. With reasoning models that think at length before answering, the output share grows further. If you only compare input prices when choosing a model, you will underestimate the real bill; always price the full request shape.