Watch: count tokens for a prompt
No sound. Each step is captioned on screen and also written out in the guide below.
AI Tools
AI Token Counter
See how many tokens your prompt uses for GPT, Claude, and Gemini, and whether it fits each model's context window.
Count tokens locally
Runs locally in your browser. Your text is counted on this device and never uploaded.
Counts update as you type. Up to 5 MB.
- Characters
- 0
- Words
- 0
- Lines
- 0
- UTF-8 bytes
- 0
Loading the tokenizer (about 1.6 MB, once)…
Context window fit
| Model | Context window | Your text | Used |
|---|
See the tokens
Paste text to see how it splits into tokens.
Each highlighted block is one token. Blocks that hold a split emoji or character count as several.
Count tokens in three steps
- 01Paste your text
A prompt, document, code, or a whole chat transcript. Nothing is uploaded.
- 02Read the counts
OpenAI tokenizers are counted exactly. Claude, Gemini, and GPT-6 are labeled estimates.
- 03Check the fit
See how much of each model's context window your text would use.
Why counts differ by model
Text o200k_base cl100k_base
日本語 2 tokens 4 tokens
👋🏽 4 tokens 6 tokens
Supercalifragilisticexpialidocious
10 tokens 11 tokensEach tokenizer splits text with its own vocabulary, so the same prompt uses a different number of tokens on each model. Non-English text and emoji differ the most.
Frequently asked questions
Which counts are exact?
The two OpenAI tokenizers, o200k_base (GPT-5, GPT-4.1, GPT-4o, o-series) and cl100k_base (GPT-4, GPT-3.5 Turbo), are counted with OpenAI's published vocabulary and match OpenAI's own tiktoken library token for token.
Why are Claude and Gemini estimates?
Anthropic and Google don't publish their tokenizers for offline use. Exact counts come from their APIs: Claude's count_tokens endpoint and Gemini's countTokens. The estimates here follow each provider's own guidance and say how they're calculated.
Why is GPT-6 an estimate?
OpenAI hasn't published GPT-6's tokenizer, and OpenAI's tiktoken library doesn't support it yet. The page shows the o200k_base count as a guide until it does.
Is my text sent anywhere?
No. The page downloads the tokenizer vocabulary from this site once, then counts on your device. Your text never leaves the browser tab.
How current are the model limits?
Context windows come from each provider's documentation, checked on the date shown under the table, with links to the sources. Providers change models often, so check the source before relying on a limit.