LLM Token Counter & Cost Estimator

A token is a sub-word chunk chosen by the model's BPE tokenizer. Common words are often a single token, but longer or rarer words, punctuation, whitespace, and non-English text can split into several tokens. A rough rule of thumb for English is ~4 characters or ~0.75 words per token, but this tool counts the actual tokens rather than estimating from length.

For OpenAI models, yes — they use OpenAI's published o200k_base and cl100k_base tiktoken encodings, so the count matches what the API bills. For Claude and Gemini the count is an approximation: those tokenizers are proprietary and aren't available as a browser library, so we tokenize with o200k_base and label the result approx. It's a good ballpark, not an exact bill.

Input cost = tokens × the selected model's input price ÷ 1,000,000. The tool shows the per-1M input and output prices it used and the date of the embedded pricing snapshot. Model prices change often, so always confirm the current rate on the provider's pricing page before budgeting. The estimate covers the input (prompt) tokens only — the model's completion is billed separately at the output rate shown.

No. It counts the plain encoding of the text you paste — the answer to "how many tokens is this text". Chat APIs add a few tokens per message for role and formatting metadata, and that overhead is provider- and endpoint-specific, so it isn't included here.

No. The tokenizer runs entirely in your browser as WebAssembly. Your text never leaves your device and no network request is made, so it's safe to paste sensitive prompts.