Tokens are chunks of text (1-6 characters each) that language models read. Common words are often single tokens; rare words and code are split into multiple. Rule of thumb: 1 token is approximately 4 characters or 0.75 English words.

Is my prompt sent anywhere?

No. All tokenization runs in your browser via js-tiktoken and gpt-tokenizer. No network requests contain your text. Safe for proprietary prompts and sensitive data.

Why do different models show different token counts?

Each model family uses its own tokenizer. Only OpenAI publishes an exact browser-compatible tokenizer. Claude, Gemini, and Llama are approximated using the OpenAI tokenizer — typically within ±10% for English text.

Are the prices accurate?

Prices reflect publicly listed rates per million tokens at the last update. Volume discounts, Batch API, prompt caching, and enterprise contracts change real costs. Always verify against the provider pricing page.

Why is output more expensive than input?

Generating a token requires a full forward pass through the model, while reading input is cheaper. Most providers price output at 3-5x the input rate. Reasoning models like o1 also charge for hidden reasoning tokens not shown in the response.

AI & LLM

Token Counter

Count tokens and estimate API costs across 100+ models from OpenAI, Anthropic, Google, Meta, Mistral, Cohere, Qwen, xAI, DeepSeek, and Perplexity. Paste any prompt, compare per-million-token pricing side by side, add custom models, and export results as JSON or CSV. 100% client-side via js-tiktoken and gpt-tokenizer — no data leaves your browser.

Open tool

Quick reference

Category: AI & LLM
Runs in: Browser
Data stored: Nothing
Price: Free
Sign-up: Not needed

Browse categories

All features

100+ models from 9 providers (OpenAI, Anthropic, Google, Meta, Mistral, Cohere, Qwen, xAI, DeepSeek)
Exact token counts for OpenAI (o200k_base, cl100k_base) and Llama (gpt-tokenizer)
Per-million-token input/output/cached pricing comparison
Output token cost estimator with live total calculation
Context window usage percentage per model
Custom model definitions with own pricing, tokenizer, and context (localStorage)
Quick stats bar (chars, words, lines, token range, tok/char ratio)
Export results as JSON or CSV

What you can do

Count tokens before API calls

Paste any prompt to see the exact token count for GPT-4o, Claude, Gemini, and Llama — avoiding unexpected context window overflow.

Estimate API costs

Enter your expected output tokens to see the estimated cost per model, so you can pick the most cost-effective option for your task.

Compare model pricing

View per-million-token input and output pricing side by side across all supported models to inform your model selection.

Add custom models

Define custom models with your own pricing, tokenizer family, and context window — saved to localStorage for persistent use.

Debug context limits

Check whether your prompt fits within a model context window before sending it, and see how many tokens remain for the response.

Private by design

Client-side
No sign-up
Free

Token Counter runs entirely in your browser tab. No data is uploaded, no server processes your input, and no account is required.

Specifications

Tokenizer: js-tiktoken (o200k_base, cl100k_base), gpt-tokenizer (Llama)
Models: 100+ from OpenAI, Anthropic, Google, Meta, Mistral, Cohere, Qwen, xAI, DeepSeek
Accuracy: OpenAI exact; Claude/Gemini/Llama ±10% approximated
Cost data: Per-million-token input/output/cached pricing
Custom models: Yes — localStorage-persisted with own tokenizer and pricing
Export: Copy JSON, download CSV
Processing: 100% client-side, no upload