How does Claude's tokenizer differ from GPT's?

Claude uses its own byte-pair vocabulary that differs from OpenAI's tiktoken. In practice Claude tends to use slightly more tokens than GPT-4o for the same English text — roughly 3.6-3.8 characters per token — which is why counts won't match a GPT counter exactly.

Can I paste a full conversation?

Yes. Paste a JSON array of messages like [{"role":"user","content":"…"}]. The tool counts each message separately and adds a small per-message overhead to approximate the role/structure tokens the API adds.

How accurate is the estimate?

It is a heuristic based on Claude's measured average characters-per-token ratio, not the official tokenizer. For English prose it is usually within 5-10%. For an exact count, call Anthropic's count_tokens endpoint.

Is anything sent to Anthropic?

No. Everything runs in your browser. Nothing you paste is uploaded, stored, or logged.

Claude Token Counter

Claude token counter

Estimate how many tokens your text or conversation will use on Anthropic’s Claude models — Opus, Sonnet, and Haiku. Token counts drive both billing and the 200K context window, so checking before you send keeps long documents and multi-turn chats inside the limit.

How the estimate works

Claude uses its own byte-pair encoding vocabulary, distinct from OpenAI’s tiktoken. This tool applies Claude’s measured average of roughly 3.7 characters per token for English, blended with a word-boundary heuristic. When you paste a JSON array of messages, each message is counted separately and a small fixed overhead per message approximates the role and structural tokens the Messages API adds around each turn.

Tips and notes

Claude usually counts slightly more tokens than GPT-4o for identical English text, so don’t reuse a GPT estimate as-is.
A multi-turn conversation accumulates: every prior turn you resend counts again toward the context window and the bill.
For exact counts in production, use Anthropic’s count_tokens endpoint, which returns the precise token total for a given request.