Count characters, words, and tokens for any prompt
Whether you are fitting a prompt into a context window, budgeting API cost, or staying under a UI character limit, you need to know how big your prompt is. This tool counts characters, words, sentences, and estimated tokens live as you type, and breaks the estimate down by system / user / assistant block. It runs entirely in your browser.
How it works
Characters, words, and sentences are exact counts. Tokens are estimated,
because the real tokenizers (GPT’s tiktoken, Claude’s proprietary tokenizer)
are large and model-specific. The estimate blends two signals:
tokens ≈ max( characters / chars_per_token, words × words_per_token )The GPT and Claude presets use slightly different ratios because their tokenizers segment text differently. For ordinary English prose the estimate is typically within a few percent of the true count; code, JSON, and other languages can drift further.
When your prompt uses role labels (System:, User:, Assistant:), each block
is counted separately so you can see exactly where your token budget is going.
Tips and notes
- A rough rule of thumb: 1 token ≈ 4 characters ≈ 0.75 words in English.
- Long system prompts are a silent cost: they are re-sent on every turn, so a bloated system block multiplies across a conversation.
- For exact billing, confirm against your provider’s usage dashboard — this tool is for fast estimates while drafting.