LLM Token Estimator.
Paste or upload the text you feed to an LLM agent and get an instant estimate of how many tokens it will use for Claude, GPT-4o, and other models, plus live character, word, sentence, and line counts. Every LLM tokenizes differently and some don't publish an offline tokenizer, so these are well-calibrated estimates, computed entirely in your browser.
// token estimates estimate
// text analytics
// how it works & the fine print
Tokens are the sub-word chunks a language model actually reads, roughly a short word or a piece of one. Pricing, context windows, and rate limits are all measured in tokens, so knowing the token size of a prompt before you send it helps you budget cost and stay inside a model's window.
Why these are estimates. Every model family uses its own
tokenizer. OpenAI publishes theirs (BPE encodings like cl100k and
o200k), but Anthropic does not ship an offline tokenizer
for current Claude models; an exact count only comes from their
count_tokens API, which would mean sending your text to a server.
Because this tool is browser-only and uploads nothing, it estimates instead:
it derives tokens from your character and word counts using each family's
typical characters-per-token ratio.
How good is the estimate? For ordinary English prose the figures are usually within about 10–20% of the real count. Source code, dense punctuation, emoji, and non-English scripts tokenize less predictably and can drift further, so treat the output as a planning guide, not an invoice. When exact numbers matter, verify against the provider's own tokenizer or API.
Ratios used: Claude ≈ 3.5 chars/token, GPT-4o ≈ 4.1, GPT-4 / GPT-3.5 ≈ 4.0, and a word-based cross-check at roughly 1.33 tokens per word. These are community-standard rules of thumb, not guarantees.
Privacy: everything (counting and estimating) runs in this browser tab with no dependencies and no network calls. Watch the Network tab: it stays quiet as you type.
Related: working with structured prompts? The JSON Formatter & Validator tidies JSON payloads, and the Markdown to PDF tool renders prompt docs.
// a free, browser-only tool by Tiempo Software, built in Omaha, Nebraska. Runs entirely in your browser; nothing you enter is uploaded. see all tools.