Skip to content
Kordu Tools
Developer Tools Runs in browser Updated 01 Apr 2026

AI Token Counter

Count tokens for GPT-4o, Claude, Gemini, and more. Exact tiktoken counts for OpenAI models, estimates for others, with API cost breakdown.

▾
o200k_base

0

Tokens

0

Characters

0

Words

$0.00

Est. Input Cost

Cost Breakdown
Input$0.00

$2.50 / 1M tokens

Output$0.00

$10.00 / 1M tokens

Pricing based on publicly available API rates — verify with the provider before production use.
Loading rating…

How to use AI Token Counter

  1. Select your AI model

    Choose the model you are targeting from the dropdown. Models are grouped by provider: OpenAI (GPT-4o, o1, o3, GPT-4.1), Anthropic (Claude 3.5, Claude 4), and Google (Gemini 1.5, Gemini 2.0).

  2. Paste or type your text

    Paste your prompt, document, or any text into the input area. Token count, word count, and character count update automatically as you type.

  3. Check context window usage

    The context bar shows what percentage of the selected model's context window your text uses — helpful for staying within limits on large documents.

  4. Review the cost estimate

    The cost panel shows estimated input and output API costs based on the model's current public pricing per million tokens.

AI Token Counter FAQ

What is a token in AI models?
A token is a chunk of text — roughly 3–4 characters or about 0.75 words in English. Tokenizers like BPE (Byte Pair Encoding) split text into common subword units. 'hello' is one token, 'chatbot' might be two. Code and non-English text typically use more tokens per word than English prose.

Is this accurate for GPT-4o and GPT-4.1?
Yes. GPT-4o, GPT-4.1, and the o-series (o1, o3) use o200k_base encoding. GPT-4 and GPT-3.5-turbo use cl100k_base. This tool uses js-tiktoken with the same rank data, so counts match OpenAI's API exactly.

How accurate are Claude and Gemini token counts?
Approximate. Anthropic and Google use proprietary tokenizers that are not publicly available. This tool uses cl100k_base as a proxy — typically within 5–10% for English text. For precise billing figures, check your provider's usage dashboard.

Why does token count matter?
Two reasons: context limits and cost. Every model has a maximum context window — exceed it and the model truncates your input or returns an error. On paid APIs, you pay per token in and out. Knowing your count upfront lets you optimise prompts and control costs before sending.

What is the context window percentage?
It shows what fraction of the model's total context window your text occupies. For example, GPT-4o has a 128K token context window. If your text is 12,800 tokens, the bar shows 10%. This helps you see how much room remains for the model's response.

Is my text sent to a server?
No. All tokenization runs entirely in your browser using JavaScript. The BPE rank data (~2–4 MB depending on encoding) is downloaded once from a CDN and cached locally. Your text never leaves your device.

Why is the first count slightly slower?
The tokenizer's BPE rank data is downloaded the first time you use the tool (~2–4 MB). After that, it is cached in your browser and all subsequent counts are instant.

Does token count include system prompts?
This tool counts the tokens in whatever text you paste. If your use case includes a system prompt, you should paste the full combined text (system prompt + user message) to get an accurate total count.

Can I use this to estimate API costs?
Yes. The cost estimate panel shows estimated input and output costs for the selected model based on current public pricing. These are estimates — actual costs depend on your provider tier, batching, caching, and any discounts in your contract.

Background

AI Token Counter tells you exactly how many tokens your text will consume before you send it to an AI API. Token count determines context window usage and API cost — both of which directly affect what you can build and what you pay.

The tool uses js-tiktoken, a JavaScript port of OpenAI's official tiktoken library, with the same BPE rank data. GPT-4o, GPT-4.1, and the o-series models use o200k_base encoding. GPT-4 and GPT-3.5-turbo use cl100k_base. Counts for these models are exact — identical to what OpenAI's API charges you for.

For Anthropic (Claude 3.5 Sonnet, Claude 3 Opus, Claude 4) and Google (Gemini 1.5, Gemini 2.0) models, cl100k_base is used as a close proxy. These providers use proprietary tokenizers, so the estimate is typically within 5–10% for English text.

The stats bar shows token count, word count, character count, and what percentage of the selected model's context window the text consumes. The cost panel breaks down estimated input and output costs based on current public API pricing per million tokens. The tokenizer loads lazily on first use and caches in memory — subsequent counts are instant. All processing is client-side.

Related tools

JSON Formatter

Format, validate, and minify JSON instantly — with configurable indentation, error location, and tree view.

Base64 Encoder/Decoder

Encode text or files to Base64 or decode Base64 strings back to plain text — real-time, fully in your browser.

Hash Generator

Generate MD5, SHA-1, SHA-256, and SHA-512 hashes from text or files instantly in your browser.

Word Counter

Count words, characters, sentences, and paragraphs with reading time, speaking time, and keyword density.

Learn more

developer

Character Encoding Explained: UTF-8, ASCII, Unicode

UTF-8, ASCII, and Unicode explained with byte-level examples. Learn why mojibake happens, how to fix encoding issues, and why UTF-8 powers 98.2% of the web.