ChatGPT Token Counter, Also for Claude, Gemini and Grok

Roughly 1,333 tokens on OpenAI, Google, xAI and Claude Haiku 4.5, and about 1,800 on Claude Opus 5, Sonnet 5 and Fable 5.1, which use Anthropic's newer tokenizer. It runs the opposite way to how people usually remember it: one token is about three quarters of a word on most models, so a word costs slightly more than one token, not less. The exact figure moves with word length, punctuation and special characters.

About 750,000 words on OpenAI, Google and xAI models, or roughly a 1,500 page book. On Claude Opus 5, Sonnet 5 and Fable 5.1 the same 1 million tokens carries about 555,000 words, because Anthropic's newer tokenizer, introduced with Claude Opus 4.7, splits text into roughly 30 percent more tokens. Claude Haiku 4.5 predates that change, so its 200,000 token window holds about 150,000 words.

It is an estimate, not a tokenizer. It divides your word count by the average words-per-token ratio for the model you pick, which lands within roughly 10 to 15 percent for ordinary English prose. Expect more drift on code, on languages other than English, and on text heavy with punctuation, numbers or emoji, all of which tokenize less efficiently than plain words. For an exact figure use OpenAI's tiktoken library, the Anthropic tokenizer API, or xAI's tokenize-text endpoint.

No. The counting runs entirely in your browser with plain JavaScript. Your text is never sent to our servers, never stored and never logged. You can disconnect from the network and the counter still works.

The context window is the total number of tokens a model can hold at once, covering the conversation so far plus the reply it is about to write. When a conversation exceeds it, the oldest turns are dropped, which is why a long chat starts forgetting what you said at the beginning.

Because the context window is finite. Once the thread passes the limit, older messages fall out of the model's view. The Context Meter in AI Toolbox shows how full the window is and roughly how many messages you have left, and can summarise the thread into a fresh chat before anything is lost. The gauge is free on ChatGPT, Gemini, Claude and Grok.

Keep prompts concise, avoid re-pasting context the model already has, put persistent rules in custom instructions rather than repeating them, and start a new conversation for a new task. Saved prompts help too: AI Toolbox's prompt library stores reusable prompts you insert by typing two slashes, so you are not rewriting the same long preamble each time.

Slightly, because they use different heuristics. This page divides words by an average words-per-token ratio. The extension's Writing Analytics panel estimates from character count instead, at about four characters per token for Latin scripts. On a long real transcript the two landed about four percent apart. Both are estimates, so treat either as a close guide rather than an exact tokenizer count.

The current OpenAI, Anthropic, Google and xAI frontier lineups, including GPT-6 Astra, GPT-5.6 Sol, Terra and Luna, GPT-5.5 and GPT-5.4, Claude Fable 5.1, Opus 5, Sonnet 5 and Haiku 4.5, Gemini 3.1 Pro and 3 Flash, and Grok 4.5, 4.3 and 4.20. The table below lists each one's context window, maximum output and published API price.

Yes, with no account and no sign-up. The AI Toolbox extension is also free to install: the word and character line under the ChatGPT composer and the context gauge on all four platforms are on the free plan. The Writing Analytics panel with token estimates is Premium at $9.99/month or $99 once.