Runs in your browser: nothing you enter or choose is uploaded.
This AI token counter shows how many characters and words your text has and estimates how many tokens it will use, at roughly 4 characters per token for English. It also shows how much of an 8K, 32K, 128K, 200K or 1M-token context window the text would fill. It runs in your browser: nothing is uploaded.
How to use the AI token counter
- Paste your text into the Paste your text or prompt box. It can be a prompt, an email, an essay or a whole document you plan to give a chatbot.
- Watch the numbers change as you type or paste. There’s no button to press: the counts update on their own.
- Read the estimated tokens figure and the bars that show how much of each context window (8K, 32K, 128K, 200K and 1M tokens) your text would use.
- If you need an exact figure for one specific model, check it with that company’s own tool. We list them under Exact token counts for ChatGPT, Claude and Gemini below.
For more guides on chatbots and how to use them well, see our AI hub.
What is a token?
A token is a chunk of text that a language model reads or writes. A short, common word is often a single token. A long or unusual word is split into several pieces, and punctuation marks are usually tokens of their own.
Everything an AI model does is measured in tokens: how much text it can take in at once (its context window), how long its answer can be and, for developers using an API, what each request costs. That’s why it helps to know roughly how big your text is before you paste it.
How the token estimate works
The tool counts characters, including spaces and punctuation, and words, meaning runs of text separated by spaces. It then divides the number of characters by 4 to estimate tokens.
That’s the rule of thumb both OpenAI and Google publish for English. OpenAI’s help centre says one token is about 4 characters, or about three-quarters of a word, so 100 tokens is roughly 75 words. Google says a Gemini token is about 4 characters and that 100 tokens is about 60 to 80 English words.
Why not give an exact count? Because there isn’t one exact count. Each company splits text with its own tokenizer, and they change it between model generations. Anthropic’s documentation says Claude Opus 4.7 and later models use a newer tokenizer that produces roughly 30 percent more tokens for the same text than earlier Claude models. A counter that claimed to be exact for every AI would be wrong for most of them.
The window bars are simple arithmetic: estimated tokens divided by the window size, times 100. (8K means 8,192 tokens, 32K 32,768, 128K 128,000, 200K 200,000 and 1M 1,000,000.)
| Text | Words | Estimated tokens (English) |
|---|---|---|
| A short message | 40 | about 50–55 |
| An email or essay section | 500 | about 650–700 |
| A long blog post | 1,500 | about 2,000 |
| A report or long chapter | 6,000 | about 8,000 |
| A full-length novel | 90,000 | about 120,000 |
How to read your result
A context window holds everything in the conversation: your text, any earlier messages and the model’s reply. Anthropic’s documentation says this directly: the window includes the response itself. So leave room for the answer, especially if you want a long one.
- Under 25% of a window: plenty of room for your text, a follow-up conversation and a long answer.
- 25% to 75%: fine for a question or two about the text, but a long back-and-forth chat will run out of space sooner. Start a new chat for a new task.
- Over 75%: answers may be cut short, and in a long chat the model may lose track of earlier messages. Split the text, summarise it in parts, or use a model with a bigger window.
- Over 100%: it won’t fit. Through the API, Anthropic rejects a request whose input is already larger than the window with a “prompt is too long” error.
Bigger isn’t automatically better. Anthropic notes that accuracy and recall drop as the token count grows, which it calls “context rot”. Paste the parts that matter rather than everything you have.
Exact token counts for ChatGPT, Claude and Gemini
OpenAI (ChatGPT and the API)
OpenAI has a free Tokenizer page at platform.openai.com/tokenizer, where you paste text and see exactly how it’s split for its models. Developers can count tokens in code with OpenAI’s open-source tiktoken library by choosing the encoding for their model.
Claude (Anthropic)
Anthropic documents token counting in its API docs. The Messages API has a token counting endpoint that returns the input tokens for the model you name. It’s free to use, with a rate limit, and Anthropic itself calls the result an estimate that can differ by a small amount. At the time of writing (October 2026), Anthropic lists a 1M-token context window for its current models, such as Claude Opus 5.5 and Claude Sonnet 5.5, and 200K for older ones such as Claude Sonnet 4.5.
Gemini (Google)
The Gemini API has a count_tokens call that counts your input before you send it, and every response reports the tokens it used. Google’s guide also explains that images, audio and video cost tokens: a small image counts as 258 tokens, for example. Each model’s input limit is listed on its page in Google’s documentation.
In the ChatGPT, Claude and Gemini apps you won’t see a token count at all. The app manages the window for you, and plan limits are usually described in messages or usage rather than tokens. Our ChatGPT Free vs Plus comparison and guide to choosing an AI assistant cover what each plan gives you.
Why the same text gives different token counts
- Language: the 4-characters rule is for English. Other languages often need more tokens for the same meaning, so for non-English text treat the estimate as a likely minimum.
- Code and data: code, JSON, web addresses, long numbers and unusual spellings break into more pieces than ordinary prose.
- Formatting: extra spaces, tables and Markdown symbols are text too, and they all count.
- Model generation: a newer model from the same company can count the same text differently, as Anthropic’s tokenizer change shows.
- Hidden extras: system instructions, tool definitions, the earlier chat history and any images or files you attach all use tokens as well.
Tips to use fewer tokens
- Cut boilerplate before pasting: email signatures, repeated disclaimers and navigation text add tokens without adding meaning.
- Paste only the section you need help with, not the whole document.
- For very long documents, ask for a summary of each part, then work from the summaries.
- Start a new chat for a new topic. In a chat, every new turn sends the whole earlier conversation again, so long chats fill the window fast.
- Ask for the length you want, such as “answer in 150 words”, so the reply doesn’t use space you need.
Counting here is private, but pasting text into a chatbot isn’t always. Before you paste anything personal or confidential, check what each service keeps with our AI privacy settings guide.
Limitations
- It’s an estimate, not an exact count. It can be noticeably off for code, tables and non-English text.
- It counts text only. Images, PDFs and audio you attach to a chat use tokens that this tool can’t see.
- The window sizes are reference points. Check the real limit for the model and plan you use, and remember the reply shares the same space.
- It doesn’t estimate cost. API prices differ by model and by input and output tokens, so use each company’s official pricing page.
FAQ
How many tokens is 1,000 words?
Roughly 1,300 to 1,400 tokens of English, using OpenAI’s rule of thumb that 100 tokens is about 75 words. Google’s figure of 60 to 80 words per 100 tokens gives a range of about 1,250 to 1,670. The exact number depends on the model.
Is a token the same as a word?
No. Common short words are often one token, but longer or rarer words are split into several, and punctuation is usually separate. On average a token is about three-quarters of an English word.
How do I count tokens for ChatGPT?
For a quick estimate, paste your text into this tool. For an exact count with OpenAI’s models, use OpenAI’s Tokenizer page, or the tiktoken library if you’re writing code.
Does the AI’s answer count toward the context window?
Yes. The window holds your input, the earlier conversation and the model’s reply together, so a prompt that fills most of the window leaves little room for the answer.
Why do Claude and ChatGPT give different token counts for the same text?
Each company uses its own tokenizer, which splits text into different pieces. Counts can even change between generations from the same company, so always check against the model you plan to use.