Count tokens and estimate cost across 15+ AI models
Paste a prompt. Get instant token counts and per-million-call cost for Claude, GPT-4o, Gemini, Llama, and more. Free, no signup, no logging.
| Model | Tokens | Confidence | Per call | Per 1M calls |
|---|---|---|---|---|
| Type or paste a prompt to see counts. | ||||
What is an AI token?
A token is the smallest unit a language model reads. Roughly, 1,000 tokens is about 750 words of English — but the exact split depends on the model's tokenizer. This is why a 10,000-character prompt can cost 30% more on one model than another: different vocabularies break the same text into different numbers of pieces.
Why we built this
Every provider has its own tokenizer. OpenAI publishes tiktoken. Anthropic exposes count_tokens in its API. Google has countTokens. Open-source models each ship their own. To compare cost honestly, you need the right tokenizer for each model — not one heuristic stretched across all of them.
We do that. Our results page shows a confidence label per model (exact, ≈±3%) so you always know what you're looking at. Read the methodology for the full breakdown.
Frequently asked
How accurate are the token counts?
OpenAI, Anthropic, and Gemini counts are exact — they use each provider's official tokenizer. Llama, Mistral, DeepSeek, and Qwen counts are approximations within ±3% of the reference tokenizer. Every row in the results table shows its own confidence label.
Is my prompt sent anywhere?
OpenAI counts run entirely in your browser. Anthropic and Gemini counts go through our serverless proxy to those providers' tokenization endpoints — the prompt is sent for tokenization only, never logged, never stored, never used for training.
How fresh is the pricing?
We snapshot pricing manually whenever a provider updates. The "Pricing as of" date in the counter shows the last update. The full pricing changelog documents every change.
Why does the same text cost more on one model?
Two reasons: different tokenizers (different number of tokens for the same text) and different per-token prices. Both are visible in the table above — that's the whole point.