Frequently asked questions
How Claude tokenization works, API counting, and pricing implications.
- What is a Claude tokenizer?
- A tokenizer splits text into tokens — the units Claude models read and bill. Token count depends on model and content; character count alone is not reliable.
- Is this count exact?
- Yes. We use Anthropic's official count_tokens API — the same logic used for billing and context limits.
- Do I need an API key?
- No. You use the website directly. Our server may use an API key to call Anthropic on your behalf for accurate counts.
- Which models are supported?
- Claude Opus 4.x, Sonnet 4.x, Haiku 4.x, and Claude 3.5 Sonnet. Select the same model you plan to use in production.
- How does this relate to API cost?
- Input tokens from count_tokens match what Anthropic uses for input billing (plus any system prompts or tools you add in real requests). Output tokens are counted separately at generation time.