AI Token Calculator

1. Give the tool your input size. Paste text, type a word count, or enter an exact token count. An exact count wins over text, and text wins over a word count.
2. Enter the output tokens per call and your own prices per 1M tokens from your provider price list. Add calls per day and active days per month.
3. Read the token estimate, the range and the cost. Optionally add price B to compare a second model or provider on the same usage.

Here is what each result means:

An AI token is the unit of text a language model processes. Models do not read letters or whole words. They read tokens, and providers bill by the token.

OpenAI describes tokens as commonly occurring sequences of characters. Its example splits the word "tokenization" into two tokens, "token" and "ization", while a short common word like "the" is one token. Anthropic describes tokens as the smallest units of a language model, which can be words, subwords, characters or bytes.

Both vendors publish a rule of thumb for English text, and they do not match:

These are approximations, not billing rules. Anthropic adds that the exact number varies with the language used. This tool is a planning aid. It is not a tokenizer and is not affiliated with any provider.

The calculator divides your text size by a ratio to estimate tokens, then multiplies tokens by your price per million.

1. Count the characters in the pasted text, or take the word count you typed.
2. Divide characters by the characters-per-token ratio, or words by the words-per-token ratio. Round up to a whole token.
3. Multiply input tokens by your input price and output tokens by your output price. Add them and divide by 1,000,000.
4. Multiply the cost per call by calls per day and active days per month.

Prices per million tokens are how providers quote them. Anthropic writes the unit as MTok, and OpenAI writes "per 1M tokens". At the OpenAI rule of thumb, one million tokens is about 750,000 words or 4,000,000 characters.

A prompt is 750 words long. The reply is about 400 tokens. Your provider charges $2 per 1M input tokens and $10 per 1M output tokens. You send 1,000 calls a day for 30 days.

Tokens: 750 words / 0.75 words per token = 1,000 input tokens.

Cost per call: input = 1,000 x $2 / 1,000,000 = $0.002. Output = 400 x $10 / 1,000,000 = $0.004. Together that is $0.006.

Per month: 1,000 x 30 = 30,000 calls. 30,000 x $0.006 = $180, or $2,160 a year. The month uses 30,000,000 input tokens and 12,000,000 output tokens.

Compare price B: at $0.50 input and $3 output, one call costs $0.0017 and the month costs $51. Price B is $129 cheaper, which is 71.7% less. The prices in this example are sample figures, not a quote from any provider.

The Tokenizer

Each model family has its own tokenizer, so the same text gives different counts. Anthropic states that its Claude 4.7 and later models produce about 30% more tokens for the same text than its earlier models. A prompt that counted 1,000 tokens becomes roughly 1,300.

The Language and Content Type

The rules of thumb are for English prose. Anthropic notes that the number of characters per token varies by language. Code, tables and other languages can tokenize very differently, so lower the characters-per-token value to be safe.

Hidden Parts of a Request

A real request holds more than your message. OpenAI says the count includes formatting tokens for message roles and boundaries, and that tool definitions add tokens. System instructions and chat history also count as input.

Images and Files

Character ratios do not work for images or files. OpenAI says estimates like characters divided by 4 are inaccurate there. For image work, see the AI image generation cost calculator.

Output Length

Output tokens are counted and priced separately. OpenAI notes that reported output includes all tokens the model generates, not only the visible text.

An estimate is fast and good enough for a first budget. An exact count is what you are billed on. Use the right one for the job.

Anthropic says its token counting endpoint is free to use and that its count is itself an estimate that can differ from actual usage by a small amount. The two vendor ratios alone differ by about 14%: 6,000 characters is 1,500 tokens at 4 per token and 1,715 at 3.5. Treat any ratio result as a range.

Sizing a Prompt

Check whether a long document is likely to fit a model. OpenAI notes that the prompt and the output together must stay within the model context length. File sizes are a different measure; the data storage converter handles bytes.

Drafting a Budget

Turn a word count and a call volume into a monthly figure using the prices on your own provider page. When a price changes, type the new one and read the result again.

Comparing Two Options

Enter a second price pair to see what the same usage costs on another model. The percentage calculator helps with any other percent gap, and the token generation speed simulator covers response time.

1. Treating a Ratio as a Fact

Four characters per token is a rough English average from one vendor. Another vendor says 3.5. Neither is a billing rule.

2. Counting Only the Visible Message

System instructions, tool definitions and earlier turns are all input tokens. Estimating from the user message alone understates the count.

3. Reusing a Count Across Models

A count measured on one tokenizer does not carry over to another. Recount when you change model.

4. Forgetting Output Tokens

Output has its own price, and it is often higher than the input price. In the example above the 400 output tokens cost twice as much as the 1,000 input tokens.

5. Mixing Up Price Units

Price lists quote dollars per million tokens. Entering a per-thousand price here overstates the cost a thousand times.

6. Using Old Prices

Providers change prices and retire models. Read the price on the official page on the day you budget, and note the date.

The cost arithmetic is exact. The token estimate is an approximation unless you enter an exact count.

About 1,334 tokens at the OpenAI rule of thumb of 0.75 words per token for English text. The real count depends on the model tokenizer and the text, so treat it as an estimate.

About 4 characters according to OpenAI and about 3.5 English characters according to Anthropic. Both figures are approximations for English text.

About 750,000 words at 0.75 words per token. That is the OpenAI rule of thumb for English and will differ for code or other languages.

Multiply the number of tokens by the price per million tokens and divide by 1,000,000. Do it once for input and once for output, then add the two.

No. It estimates tokens from characters or words. For an exact count, use your provider token counting endpoint or the usage numbers in an API response, then type that count into the exact tokens field.

Each model family uses its own tokenizer. Anthropic, for example, states that its newer models produce about 30% more tokens for the same text than its earlier ones.

Yes, they are part of the text that gets tokenized. OpenAI notes that the first token of each word in a sentence typically starts with a space character.

Enter the input and output prices per million tokens from your own provider official price list, read on the day you budget. This tool does not store provider prices.

It is the maximum number of tokens a model can handle in one request. OpenAI states that the prompt and the generated output combined must be no more than the model maximum context length.

No. Character ratios do not apply to images or files. OpenAI says such estimates are inaccurate and points to its token counting endpoint for those inputs.

No. The text is counted in your browser and is not sent to our servers. Save keeps your inputs only in this browser.

• Key concepts: what tokens are and the 4-character, 0.75-word rule of thumb (OpenAI documentation, read 2026-10-02).
• Counting tokens: formatting tokens, tools, images and output counts (OpenAI documentation, read 2026-10-02).
• Glossary: tokens and the 3.5-character approximation (Anthropic documentation, read 2026-10-02).
• Token counting: free endpoint, estimate note and the newer tokenizer (Anthropic documentation, read 2026-10-02).
• Pricing: the per-million-token unit (MTok) (Anthropic documentation, read 2026-10-02).