Tool
LLM token counter.
Estimate how many tokens a prompt, document, or transcript will use. Approximation good enough for scoping. Runs in your browser — your text never leaves the page.
Try a sample:
Characters
0
Words
0
Approximation only — uses 4 characters per token, which is reasonable for English. Real tokenizers vary by model. For exact counts, use the tokenizer from your model's SDK.
What a token count is actually telling you
- Every model family counts differently
- A tokenizer splits text into the units a specific model was trained on, so an exact count only exists relative to one tokenizer. This tool gives you a close approximation — fine for scoping and budgeting, not something to bill a customer on.
- English runs roughly four characters per token
- As a rule of thumb, a token is about three-quarters of a word. Code, JSON, and languages that don't use Latin script are markedly less efficient — the same content in a non-English language can cost noticeably more tokens, which matters if you serve multiple markets.
- Order of magnitude is the decision, not the digits
- Whether a document is 2,000 tokens or 200,000 changes the architecture — one fits comfortably in context, the other needs chunking and retrieval. Whether it's 2,000 or 2,150 changes nothing at all. Use this to pick an approach, not to optimise.
- Conversations re-send their history
- The number that drives cost in a chat product isn't one message, it's the whole thread replayed on every turn. A twenty-turn conversation can cost far more than twenty single messages, and it's the most common reason a pilot's bill surprises people at scale.
Need to estimate monthly cost? Use the cost estimator →
