LLM cost calculator
Pick a chat model, enter the tokens of one request and the requests a day, and read the same workload at the developer's list price and at ours. 73 models, rates from the live catalog.
Per million tokens here: $1.20 in, $0.0750 cached, $3.00 out. Anthropic's list: $2.00 in, $0.1000 cached, $10.00 out.
| Workload | At Anthropic's list price | Here | Saving |
|---|---|---|---|
| One request | $0.00900 | $0.00390 | $0.00510 (56%) |
| 100 requests a day | $0.9000 | $0.3900 | $0.5100 (56%) |
| A month of 30 days | $27.00 | $11.70 | $15.30 (56%) |
Rates per million tokens from the live catalog when this page was built, rebuilt at most every five minutes. Cached tokens bill at the cached rate, or the input rate on a model without one; cache writes are not counted. A real request is billed on the model's own token count.
Where the rates come from
Every chat model in the calculator has a list price published by its developer; both rates come from the live catalog. The full set is the LLM API price index, also as JSON and CSV.
For a bill you already have, the bill rewriter reads an OpenAI or Anthropic usage export and prices every line. To see two models side by side, use the model comparisons.
Questions
- How is the cost of an LLM request worked out?
- Each token count times its rate per million tokens, divided by a million: input at the input rate, cached input at the cached rate, output at the output rate. The three add up to the request; the request times the requests a day, times 30, is the month.
- Why are cached input tokens counted apart?
- A model with a cached rate bills the repeated part of a prompt at that rate, which is lower than the input rate. On an agent loop or a long chat most of each prompt is a repeat, so counting it at the input rate overstates the bill. A model without a cached rate bills it at the input rate, and so does the calculator.
- Is anything I type sent anywhere?
- No. The rates arrive with the page and the arithmetic runs in your browser.
- Can I get the same estimate from the API?
- Yes. POST https://api.uplinkswitch.com/v1/estimate with input_tokens and output_tokens returns the cost of that request on every model, no key needed.
- Will my bill match the estimate?
- It is an estimate at today's rates. A real request is billed on the model's own token count, and cache writes are not counted here. The console lists every request with its tokens and its charge.