Your session has been flagged for unusual activity.
Please confirm you're human to continue.
This token counter shows how many tokens your text uses in ChatGPT, Claude, Gemini and Grok, plus the rough input cost per call. Your text never leaves your device: the tokenizer runs locally.
Loading OpenAI's tokenizer… until it's ready, counts use the ~4 characters per token estimate.
Tokens
0
Characters
0
Words
0
Chars / token
—
Estimated input cost per call
October 2026 list prices per 1M input tokens. Output tokens are billed separately; see the AI cost calculator.
Large language models don't read characters or words. They read tokens: chunks of text the model learned to treat as one unit. A common short word like "the" is one token; a longer word like "tokenization" is split into a few pieces. See the glossary entry for token for a longer explanation.
The token count of your text decides three things:
Whether it fits
Every model has a maximum context window, measured in tokens.
What it costs
APIs bill per token, with separate prices for input and output.
How fast it runs
More tokens means more compute and a longer wait for the answer.
No time to paste? These are the approximations OpenAI gives for English text:
≈ 4
characters per token
≈ 0.75
words per token
100 ≈ 75
100 tokens is about 75 words
These ratios vary. Other languages often need more tokens per word than English, and code, numbers, URLs and emoji split into many small pieces. For anything close to a limit, count the real text instead of estimating.
GPT: OpenAI's tokenizer
GPT counts use js-tiktoken, the JavaScript port of OpenAI's open-source tiktoken tokenizer, with the o200k_base encoding used by GPT-4o and later models. It's the same tokenizer OpenAI recommends for counting text. API calls add a few tokens of message formatting on top.
Claude, Gemini and Grok: estimates
These providers don't ship a browser tokenizer, so the counter derives their numbers from the GPT count with a fixed ratio. Treat them as planning numbers. For exact counts, use Anthropic's token counting API or the Gemini API's countTokens method.
The most tokens each model can handle in one request, input and output combined. Check the share your text takes with the context window visualizer.
| Model | Provider | Context window |
|---|---|---|
| GPT-6 Astra | OpenAI | 1,050,000 |
| GPT-6.1 Sol | OpenAI | 1,050,000 |
| GPT-6 Luna | OpenAI | 1,050,000 |
| Claude Fable 5.1 | Anthropic | 1,000,000 |
| Claude Opus 5.5 | Anthropic | 1,000,000 |
| Claude Sonnet 5.5 | Anthropic | 1,000,000 |
| Claude Haiku 5.5 | Anthropic | 1,000,000 |
| Gemini 4 Argon | 1,048,576 | |
| Gemini 3.8 Flash | 1,048,576 | |
| Grok 4.7 | xAI | 500,000 |
| Grok 4.3 | xAI | 1,000,000 |
Tokens, from the providers' model pages, October 2026. Some plans and apps cap the window lower than the API maximum.
Paste it into the token counter above: the Tokens box updates as you type. As a rule of thumb for English, one token is about 4 characters or three-quarters of a word, so 100 tokens is roughly 75 words. Other languages, code and emoji usually need more tokens for the same amount of text.
For GPT, yes in principle: the counter uses js-tiktoken, the JavaScript port of OpenAI's open-source tiktoken tokenizer, with the o200k_base encoding (the newest encoding published in tiktoken, used by GPT-4o and later models). Real API requests add a few tokens of message formatting, and images, files and tool definitions are counted separately. If the tokenizer library can't load, the counter falls back to a 4-characters-per-token estimate and says so under the text box.
Each provider trains its own tokenizer with its own vocabulary, so the same text splits into a different number of tokens. Anthropic, Google and xAI don't ship their tokenizers as a browser library, so this counter derives their numbers from the GPT count with a fixed ratio. For exact numbers, use the provider's own counting endpoint, such as Anthropic's token counting API or the Gemini API's countTokens method.
No. Counting happens in your browser. The page downloads the tokenizer library once from a public CDN, but the text you paste is never sent, logged or stored.
Tokenizers split long or unusual words into pieces. "Unbelievable" is one word but usually several tokens. Code, numbers, emoji and non-English scripts split into even more pieces.
Yes. Everything you send is tokenized and billed: the system message, your message, earlier conversation history and any pasted documents.
Yes, and on most models they cost several times more than input tokens. This counter measures the input side only. Use the AI cost calculator to estimate input plus output spend.
In MultipleChat you can run one prompt across ChatGPT, Claude, Gemini and Grok and compare the answers side by side.
Try MultipleChat