Estimates the token count of any text and prices it across thirteen current models from Anthropic, OpenAI and Google. Set your expected output length, number of runs and how much of your input repeats, and it shows the cost per run and for the whole job.
Paste the whole thing — system prompt, context documents, the lot. That is what you pay for.
tokens (~450 words)
how many times you send it
% of input that repeats
Tokenisers differ per model, so treat every number here as an estimate within about 10%. For an exact count, use the model provider's own tokeniser.
- Input tokens
- 0
- Characters
- 0
- Words
- 0
- Per run
- 0 tokens
Cost per run, and for all runs
| Model | In $/M | Out $/M | Per run | All runs |
|---|
Use my own rates
Fill both and a "Your rate" row appears at the top of the table.
Nothing leaves your browser. Rates from providers' pricing pages and launch announcements, updated 25 September 2026; providers change them without notice, so confirm before you budget on it.
How it works
- Token estimates use characters-per-token ratios that vary by language and content type — English, CJK, Arabic, code and JSON are all different.
- Rates were read from the providers’ own pricing pages on 18 September 2026 and the date is shown on the tool.
- The cached-share field prices prompt caching properly, which is where most of the savings are on repeated calls.
- Add your own rates for a model or provider not listed.
What it does not do
Every tool on this site says this plainly, because a tool that pretends to do more than it does costs you more than it saves.
- Token counts are estimates within roughly 10%. Every model tokenises differently — use the provider’s own tokeniser for an exact figure.
- Rates change without notice. Confirm against the provider’s page before you commit a budget.
- Long-context, batch and flex tiers price differently. These are standard-tier, short-context rates.
Nothing you type here leaves your browser
There is no account, no server and no logging. The whole tool is JavaScript running on your own machine — close the tab and it is gone. You can check: open your browser’s network tab and use it.
Questions people ask
Roughly 0.75 words per token in English, but it varies with punctuation, formatting and vocabulary. Non-Latin scripts use far more tokens per character.How many tokens is a word?
Cache the stable part of your prompt, move what a smaller model can do to a smaller model, shorten the system prompt that ships with every call, and cap output length.How do I cut API costs?
They were verified on the date shown on the tool. Providers change pricing without notice, so treat it as a starting point.Are these prices current?
Prompts that go with this
-
AI Engineering
Building on models: system prompts, evals, RAG, structured output, cost control.
7 prompts
-
System Prompts
The instruction that ships with every message, written so it is still followed at message forty.
7 prompts