Token Counter
Paste any text to count its tokens, characters and words. OpenAI counts are exact, Claude and Gemini counts are clearly labeled estimates, and you can see exactly how the text gets split into tokens.
Exact. Counted with OpenAI’s open-source o200k_base tokenizer, the one GPT-4o, GPT-4.1, GPT-5 and the o-series use.
Counted in your browser. Nothing you type is sent or saved.
See the tokens
Each colored block is one token, split the way OpenAI’s o200k_base tokenizer (GPT-4o, GPT-4.1, GPT-4.5, GPT-5, o-series) splits it. A dot (·) is a space and ↵ is a new line.
How much of the context window this uses
API limits as of October 2026. Each bar uses that model’s count.
What is a token?
A token is a chunk of text that an AI model reads and writes. A short, common word is often a single token, while a longer or rarer word gets split into several pieces. Anthropic describes tokens as pieces that “can correspond to words, subwords, characters, or even bytes.” Your prompt and the model’s reply both count toward its limits and, in the API, its price.
For the full explanation, with examples, read what tokens are in AI.
Why token counts differ by model
Each AI company builds its own tokenizer, so the same text gives different counts in ChatGPT, Claude and Gemini.
- OpenAI (exact here). OpenAI publishes its tokenizers. GPT-4o, GPT-4.1, GPT-4.5, GPT-5 and the o-series use one called o200k_base. GPT-4 and GPT-3.5 use cl100k_base. This page runs the real thing in your browser.
- GPT-6 (approximate). OpenAI’s open-source tokenizer doesn’t list the GPT-6 family, so we can’t confirm which one it uses. The tool shows the o200k_base count and labels it approximate.
- Claude (estimate). Anthropic hasn’t released Claude’s tokenizer. Claude Opus 4.7 and later use a newer tokenizer that, per Anthropic, turns the same text into about 30 percent more tokens than earlier models. Anthropic’s numbers work out to about 2.5 characters per token, which is what the estimate uses. For exact counts, Anthropic offers a free token counting API.
- Gemini (estimate). Google says that for Gemini models “a token is equivalent to about 4 characters,” so the estimate divides characters by 4.
Why AI can’t count the r’s in strawberry
Click Try: strawberry in the tool. On its own, the word splits into 3 tokens for GPT-4o and GPT-5: st, raw and berry. Now click In a sentence. When you actually ask “How many r’s are in strawberry?”, the word comes with a space in front of it, and “ strawberry” is a single token. The model gets one number, not ten letters, so it never sees the three r’s directly.
Tokens aren’t the whole story, since counting repeated letters is a separate weakness. Read more in how many r’s are in strawberry.
Tokens to words: rules of thumb
These are averages for English text. Paste your real text above for an actual count.
- OpenAI: “As a rough rule of thumb, 1 token is approximately 4 characters or 0.75 words for English text.” So 100 words is about 133 tokens.
- Claude (current models): “1M tokens is roughly 555k words or 2.5M Unicode characters on the current tokenizer (introduced with Claude Opus 4.7).” Older Claude models were about 3.5 characters per token.
- Gemini: “A token is equivalent to about 4 characters. 100 tokens is equal to about 60-80 English words.”
Non-English text, code and numbers often take more tokens per character than plain English.
Context windows
A context window is the most tokens a model can handle in one request, counting your prompt and its reply. These are the API limits for current flagship models as of October 2026:
| Model | Context window (tokens) |
|---|---|
| Claude (Fable 5.1, Opus 5.5, Sonnet 5.5, Haiku 5.5) | 1,000,000 |
| GPT-6 Astra, GPT-6.1 Sol, GPT-6 Luna | 1,050,000 |
| Gemini 4 Argon | 1,048,576 |
Gemini 4 Argon’s limit is shared between input and output. Limits inside the ChatGPT and Claude apps aren’t covered here and may differ by plan. Bigger isn’t always better: Anthropic notes that as token count grows, accuracy and recall degrade.
More
Tokens also explain why chatbots slip on arithmetic. See why AI is bad at math and how to get reliable numbers anyway. For spreadsheet work, the Excel formula generator gives you formulas that do the math for you, and our guide to Claude for Excel shows how to use Claude inside your spreadsheets. New to Claude? Start with the free Claude course.
Frequently asked questions
How do I count tokens for ChatGPT?
Paste your text into the box above and pick the ChatGPT model you use. For GPT-4o, GPT-4.1, GPT-5 and the o-series the count is exact, because OpenAI publishes the tokenizer those models use (o200k_base). GPT-4 and GPT-3.5 use an older one (cl100k_base), which is also in the dropdown. For the GPT-6 family the count is approximate, because OpenAI’s open-source tokenizer doesn’t list GPT-6 yet.
Is the Claude token count exact?
No. Anthropic hasn’t released Claude’s tokenizer, so any offline Claude count is an estimate. This tool divides your character count by about 2.5, which matches Anthropic’s figure for current Claude models (roughly 2.5 million characters per 1 million tokens). For an exact number, use Anthropic’s token counting API. It’s free, but it needs an API key.
How many tokens is 1,000 words?
For ChatGPT and other OpenAI models, about 1,333 tokens. OpenAI’s rule of thumb is that 1 token is about 0.75 words of English, so you divide words by 0.75. Current Claude models produce more tokens for the same text (about 1,800 for 1,000 words, using Anthropic’s 555k words per 1M tokens). Non-English text, code and numbers often use more tokens, so paste your real text for a better number.
Why does the same text give different token counts?
Every AI company trains its own tokenizer, with its own list of word pieces. A word that is one token for GPT-5 can be two or three for another model. Even the same word changes with context: “strawberry” on its own is 3 tokens for GPT-4o, but “ strawberry” with a space in front, as it appears in a sentence, is 1 token.
Does this tool send my text anywhere?
No. The OpenAI tokenizer is downloaded to your browser the first time you use the page (about 1 MB), and all counting happens on your device. Nothing you type is sent to us or to any AI company.