AI Token Counter
Paste a prompt or document to see its character and word counts and an estimated token count. Add an expected reply length and a number of requests to see tokens per request and in total. Token figures are estimates from simple rules of thumb, not exact counts from any model's tokenizer.
Runs in your browser
Free, no sign-up. Counting runs in your browser. The text you paste is never uploaded, stored or sent to analytics.
Prompt or document
All token figures are estimates from simple heuristics (characters / 4 and words x 1.33). They are not exact counts from any model's tokenizer.
Characters
0
Words
0
Est. tokens (chars / 4)
0
Est. tokens (words x 1.33)
0
Usage planner
See how many tokens a prompt uses once you add the reply and run it many times. Useful for context-window and rate-limit planning.
Est. input tokens
0
Tokens / request (in + out)
500
Total for 1,000 requests
500,000
Total = (input tokens + output tokens) x requests. Each model tokenizes a little differently, so treat this as a planning range.
How to use the AI Token Counter
- Paste your prompt, system message or document into the text box.
- Read the characters, words and the two token estimates (characters / 4 and words x 1.33).
- Set the expected output tokens per request and the number of requests.
- Choose which input estimate to use (lower, midpoint or higher), then read tokens per request and the total.
- Press Copy summary or Download to keep the figures.
Worked example
Sample data
Text: Summarize this support ticket in two sentences. Output tokens per request: 500 Requests: 1000 Input estimate: midpoint
Characters: 47 Words: 7 Est. tokens by chars / 4: 12 Est. tokens by words x 1.33: 10 Midpoint used: 11 Tokens per request: 511 Total for 1000 requests: 511,000
Tokens per request = 11 input + 500 output = 511. Total = 511 x 1000.
How it works
Why these are estimates
Language models split text into tokens using their own tokenizer, and different models split the same text differently. This tool does not bundle a tokenizer, so it cannot give an exact count.
It shows two common rules of thumb for English text: about 4 characters per token, and about 1.33 tokens per word. Code, non-English text, numbers and emoji can differ a lot, so treat the range as a rough guide and use your provider's own token-counting tool when exactness matters.
tokens (chars) = ceil(characters / 4) tokens (words) = ceil(words x 1.33)
Planning many requests
A request uses the prompt tokens plus the reply tokens. The planner adds your expected reply length to the input estimate, then multiplies by the number of requests.
Use the total to check context-window fit and rate limits before you run a batch job.
tokens per request = input_tokens + output_tokens total = tokens per request x requests
Output tokens
The reply length is not known in advance, so you enter an expected number of output tokens per request. Long replies can easily outweigh the prompt itself.
Inputs, outputs and limits
Inputs
- Text to estimate
- Expected output tokens per request
- Number of requests
- Which input estimate to use
Outputs
- Characters, words and lines
- Two estimated token counts and a range
- Tokens per request and in total
- Copy summary and download as .txt
Limits
- Token counts are heuristic estimates, not exact tokenizer output.
- Caching, batching and non-text tokens are not modelled.
- Requests are capped at 1,000,000,000 for the calculation.
Processing
- Counting runs in your browser. The text you paste is never uploaded, stored or sent to analytics.
Frequently asked questions
How accurate is the token count?
It is an estimate. For typical English prose the two heuristics are often within roughly 10 to 20 percent of a real tokenizer, but code, other languages and unusual text can be further off. Use your provider's official counter when you need an exact number.
Is my prompt uploaded?
No. Everything runs in your browser and nothing you paste is sent anywhere, including to analytics.
Which estimate should I plan with?
Choose the higher estimate for a safe margin and the lower for an optimistic one. The midpoint is the average of the two.
Does it count the tokens in my chat history or tool definitions?
Only the text you paste. Add the history, system prompt and tool definitions you plan to send to the text box to include them.
Need an assistant connected to your own knowledge?
We build AI assistants and automations grounded in your documents and systems, with access controls and evaluation built in.