AI Token Counter

Every AI model doesn't read text as words or characters — it reads it as tokens, chunks produced by a specific tokenizer that OpenAI, Anthropic, and Google each define differently for their own models. That's why the same sentence can cost a meaningfully different number of tokens depending on which model you're asking, and why "just estimate 4 characters per token" only gets you in the right neighborhood, not an exact answer.

For OpenAI's models, this tool doesn't estimate — it runs the actual tokenizer. GPT-4o and GPT-4o mini use the o200k_base encoding; GPT-4, GPT-4 Turbo, and GPT-3.5 Turbo use cl100k_base. Both are loaded via js-tiktoken, a JavaScript port of OpenAI's own open-source tiktoken library, so the token count you see here is the same count the OpenAI API would actually charge you for.

Anthropic and Google don't publish a browser-usable tokenizer for Claude or Gemini, so those figures are clearly labeled as an estimate (roughly 4 characters per token, the standard rule-of-thumb figure for English text) rather than presented as if they were exact — a real, if unglamorous, limitation worth being upfront about rather than quietly guessing and calling it precise.

The token visualizer breaks your text into its actual token boundaries with alternating highlight colors, so you can see directly why a chunk of text costs more or fewer tokens than you'd expect — repeated words often collapse into a single token, while unusual words, code, or non-English text frequently split into several. The cost calculator deliberately doesn't pre-fill any pricing, since provider rates change often; you enter the current per-million-token price yourself and it does the math.