LLM Token Counter

Free, instant, private. Count tokens and estimate API costs across 18 LLM models — entirely in your browser. Your text never leaves your machine.

→ Try it live: llmtokens.vercel.app

---

What it does

Paste any prompt or context and instantly get:

• Exact token count for all OpenAI models (uses the same tiktoken library OpenAI uses)
• ~5% approximate count for Claude and Gemini (providers don't publish tokenizers)
• Cost estimate at current API input pricing
• All 18 models at once in the Comparison tab, sorted cheapest-first

Demo

Paste any text...

┌──────────────────────────┬────────┬────────────┬──────────────────────┐
│ Model                    │ Tokens │ $/1M input │ Cost for this text   │
├──────────────────────────┼────────┼────────────┼──────────────────────┤
│ Gemini 2.0 Flash         │ ~47    │ $0.10      │ $0.000005            │
│ GPT-4o mini              │  47    │ $0.15      │ $0.000007            │
│ Gemini 2.5 Flash         │ ~47    │ $0.30      │ $0.000014            │
│ GPT-4.1                  │  47    │ $2.00      │ $0.000094            │
│ GPT-4o                   │  47    │ $2.50      │ $0.000118            │
│ Claude Sonnet 4.6        │ ~47    │ $3.00      │ $0.000141            │
│ Claude Opus 4.7          │ ~47    │ $5.00      │ $0.000235            │
│ o1                       │  47    │ $15.00     │ $0.000706            │
└──────────────────────────┴────────┴────────────┴──────────────────────┘
                                                  ↑ cheapest highlighted

Supported models

Provider	Accuracy	Models
OpenAI	Exact	GPT-4o, GPT-4o mini, GPT-4.1, o1, o3, o3-mini, GPT-3.5 Turbo
Anthropic	~5%	Claude Opus 4.7, Claude Sonnet 4.6, Claude Haiku 4.5, Claude 3.7 Sonnet, Claude 3.5 Sonnet, Claude 3 Haiku, Claude 3 Opus
Google	~5%	Gemini 2.5 Pro, Gemini 2.5 Flash, Gemini 2.0 Flash, Gemini 1.5 Pro

Token accuracy

• OpenAI: exact count using js-tiktoken (WASM port of the official tiktoken library)
• Anthropic / Google: ~5% approximation using the GPT-4 tokenizer as a proxy — these providers don't publish their tokenizers. Accurate enough for cost budgeting and prompt sizing decisions.

No install needed

Open llmtokens.vercel.app in any browser. Works on mobile.

To run locally:

git clone https://github.com/SolvoHQ/llm-token-counter
cd llm-token-counter
npm install && npm run dev
# → http://localhost:5173

Features

• Counter tab — pick one model, see token count + cost
• Comparison tab — all 18 models ranked cheapest-first, cheapest highlighted
• Copy link — share your exact prompt as a URL (text + model encoded in URL hash, nothing sent to any server)
• Privacy by design — pure client-side WASM, zero network requests for tokenization

Stack

• React + TypeScript + Vite
• js-tiktoken — WASM tokenizer, 100% client-side
• No backend · Cookieless analytics (GoatCounter) · No cookies

---

⭐ Star this repo if it saved you time

Pricing data is manually maintained. Check each provider's pricing page for the latest rates.