Token Counter

A typical single-spaced page of English text, around 500 words, runs about 650 to 700 tokens with a modern tokenizer, since one token is usually a bit less than one word. Dense technical text with code, numbers or unusual formatting tokenizes less efficiently and can run noticeably higher.

Roughly 750 words of plain English, using the common rule of thumb that one token is about four characters or three-quarters of a word. The exact ratio shifts with vocabulary and formatting, which is why this tool counts your actual text instead of applying a fixed multiplier.

No. Anthropic and Google both use their own tokenizers and have not published them openly, so an exact Claude or Gemini count is not available client-side. This tool prices the OpenAI o200k_base count against Gemini rates as an estimate, and scales it by about 1.3 for Claude 4.7 and later, whose newer tokenizer Anthropic says produces around 30% more tokens. Both are marked as estimates rather than presented as exact figures.

A token is the unit a language model actually reads and bills by, usually a word, part of a word, or a punctuation mark, not a full word every time. "Tokenization" is common and "un-", "believ", "able" is three tokens for one uncommon word, which is why token counts and word counts diverge more on technical or unusual text.

Because the count alone answers half the question. A token count that stays a count leaves you to do the pricing math by hand, per model, per plan. This tool prices your exact text at your stated volume across Claude, GPT, Gemini and DeepSeek in the same table, using the pricing data behind the rest of the cost tools on this site.

No. Counting runs entirely in your browser using a JavaScript tokenizer; the text you paste is never transmitted to a server. That also means very long pastes, a full document or a codebase, count instantly without an upload step.