Token Counter

Paste it above and you'll get the exact count. Tokens are the chunks an AI model reads, roughly 0.75 words or ~4 characters of typical English each, but the real number depends on the exact words, punctuation, code, and language. This tool runs the actual GPT tokenizer (o200k) in your browser, so the token count is exact for GPT models, not an estimate.

Close for Gemini, and no longer close for the Claude models you would actually buy. Each model family has its own tokenizer. GPT (o200k) is shown exactly here, and Gemini plus Claude Sonnet 4.6 and earlier tokenize the same English text within roughly 10-15% of that number. Anthropic says Claude 4.7 and later use a newer tokenizer that produces about 30% more tokens for the same text than the one Claude Sonnet 4.6 and earlier use, and that is its whole current lineup. Treat the Claude figures here as a floor, and use Anthropic's token counting endpoint before you commit to a budget.

Two reasons. First, models have a context window, a maximum number of tokens they can read at once, so a long document may not fit. Second, APIs bill per token, so token count is your cost. This tool shows both: how much of each 2026 model's context window your text uses, and the input cost to send it.

No. The tokenizer runs entirely in your browser using JavaScript, so your text never leaves your device and nothing is uploaded, logged, or stored. You can safely paste private prompts, documents, or code.

Trim redundant context, remove boilerplate and repeated instructions, summarize long background instead of pasting it whole, and for repeated queries over a large corpus use retrieval (RAG) so you only send the relevant chunks. Tightening a prompt usually cuts both cost and latency without hurting quality.