Claude Token Counter
Anthropic doesn't ship its Claude tokenizer as an offline library, so TokenMath counts Claude tokens with a calibrated estimate tested in our public tokenizer accuracy benchmark. Expect results within about ±3% for prose and code.
Universal AI Token Counter
Zero-latency, client-side BPE tokenization with real-time multi-model benchmarks.
Claude Models: Context Windows & Token Prices
| Model | Context Window | Input / 1M | Output / 1M | Token Count Accuracy |
|---|---|---|---|---|
| Claude Fable 5.1 | 1M tokens | $10 | $50 | Calibrated Claude Tokenizer (±3%) |
| Claude Opus 5 | 1M tokens | $5 | $25 | Calibrated Claude Tokenizer (±3%) |
| Claude Sonnet 5 | 1M tokens | $2 | $10 | Calibrated Claude Tokenizer (±3%) |
| Claude Haiku 4.5 | 200k tokens | $1 | $5 | Calibrated Claude Tokenizer (±3%) |
How Claude Tokenization Works
Claude models split text with the Claude tokenizer tokenizer, a BPE with byte fallback scheme with a ~65,000 tokens vocabulary. Common words map to one token, while rare words, identifiers and non-Latin scripts are split into several subword pieces. Read how BPE and SentencePiece tokenization work for the details.
- Claude's smaller vocabulary splits code and rare words slightly more than OpenAI's o200k_base, typically 5–9% more tokens for the same text.
- Prompt caching can discount repeated system prompts and documents by up to 90%.
- For billing-exact numbers, the Anthropic API's count_tokens endpoint is the source of truth.
Once you know your token count, estimate a full monthly bill with the LLM cost calculator, or see every Anthropic model price.
Claude Token Counter FAQ
How do I count Claude tokens?
Paste your prompt into the counter above with a Claude model selected. TokenMath tokenizes the text in your browser and shows the token count, context window usage and input cost. Accuracy: Calibrated Claude Tokenizer (±3%).
What tokenizer does Claude use?
Claude models are counted with the Claude tokenizer tokenizer (BPE with byte fallback, ~65,000 tokens vocabulary). Anthropic doesn't ship its Claude tokenizer as an offline library, so TokenMath counts Claude tokens with a calibrated estimate tested in our public tokenizer accuracy benchmark.
How many words is 1,000 Claude tokens?
For typical English prose, 1,000 tokens is roughly 700–750 words. Code, JSON and non-English text use more tokens per word, so count the actual text when budgeting.
How much does Claude Sonnet 5 cost per token?
Claude Sonnet 5 costs $2 per 1M input tokens and $10 per 1M output tokens. That's $0.00200 per 1,000 input tokens.