About

tokencalculator.net exists because most token calculators do not calculate. They estimate, usually characters divided by four, and print a model name next to the result.

The rule this site runs on

Every number is either exact or labeled. Exact means one of two things: the provider's real tokenizer running in your browser (OpenAI, Kimi, DeepSeek, all open source), or the provider's own counting endpoint (Claude and Gemini, which publish no tokenizers). Where neither exists, we say estimate, in words, next to the number.

Verification

Counters are checked against provider counting APIs on a fixed 100-prompt corpus covering prose, code, JSON, and multilingual text. The corpus re-runs when providers ship new models, which is how tokenizer changes between model generations become measurable instead of anecdotal. Methods and current status: how we count.

What's here

• Token counters for Claude, OpenAI, Gemini, Kimi, and DeepSeek
• API pricing for every major provider, with a cost calculator that models cached input
• Context windows for current models, with a fit checker
• Guides to tokens, counting, and prompt caching

Data on this site carries a verification date. If a provider changes prices or ships a new tokenizer, the date tells you how fresh the numbers are.