How many tokens
is that?
Paste any text for exact GPT token counts — computed locally with the real open-source tokenizer — plus Claude and Gemini estimates.
How do I count ChatGPT tokens? Paste your text below. GPT-4o / GPT-5-class counts are exact (o200k_base tokenizer, run in your browser); GPT-4 / 3.5 counts are exact too (cl100k_base). Claude and Gemini columns are estimates — their tokenizers aren't public. A 500-word page ≈ 650–700 tokens.
Token counter
Exact columns use the official open-source tokenizers (tiktoken o200k_base / cl100k_base) running in your tab — your text is never uploaded. Estimates assume English; CJK and code can differ by more.
Token counts by model family
| Column | Tokenizer | Accuracy |
|---|---|---|
| GPT-4o, GPT-4.1, GPT-5-class | o200k_base | Exact — runs locally |
| GPT-4, GPT-3.5 | cl100k_base | Exact — runs locally |
| Claude (Sonnet / Opus / Haiku) | Not public | Estimate, ±15% typical |
| Gemini (2.x / 3.x) | Not public | Estimate, ±15% typical |
Why token counts matter
- Cost — APIs bill per token; a long prompt sent daily adds up fast
- Context limits — every model has a window (commonly 128k tokens); going over means truncation or errors
- Prompt efficiency — trimming boilerplate from a system prompt saves tokens on every single call
Rules of thumb
English: 1 token ≈ ¾ of a word ≈ 4 characters. Code and JSON cost more per character (symbols split). CJK text costs roughly 1 token per character. A 500-word article lands around 650–700 tokens; a full book page of dense text around 500–600.
How many tokens is my text?
Paste it here and you'll get exact GPT token counts (o200k_base for GPT-4o and newer, cl100k_base for GPT-4 and 3.5) plus estimates for Claude and Gemini. A typical 500-word page is roughly 650–700 tokens.
Are the Claude and Gemini counts exact?
No — Anthropic and Google don't publish their tokenizers, so those columns are estimates (±15% typical). The OpenAI counts run the actual open-source tokenizer locally, so they are exact.
Is my text uploaded?
Never. The tokenizer (about 2.4 MB) downloads once and runs in your browser; your text stays in the tab.
What is a token?
A token is a chunk of text the model reads — roughly ¾ of a word in English. Punctuation, spaces, and code symbols each cost tokens too, which is why character counts and token counts never match exactly.