A token is a unit of text that an LLM processes—roughly a word or part of a word. Token counts vary by model and language. This calculator estimates tokens from your text so you can stay within context limits and estimate API cost.

Why do token estimates differ by model family?

Different models use different tokenizers. GPT-4 and Claude typically use about 1 token per 4 characters for English; smaller models like Llama or Mistral can use more tokens per word. Choosing a model family gives a more accurate estimate for that API.

What are special tokens?

Special tokens are added by the API for chat or completion formatting (e.g. role labels, end-of-turn markers). Including them in the estimate gives a closer count to what the API will charge for a full request.

How do I use this for API cost estimation?

Multiply your estimated token count by the model’s per-token price (input and output). Use the PDF Token Size Estimator if you need token counts for whole PDFs or RAG chunking.

Is the estimate exact?

No. The calculator uses approximate rules per model family. Actual tokenization depends on the exact model and API. Use it for planning and budgeting; check your provider’s usage for precise counts.

Text to Token Calculator

Paste or type text to see character count, word count, and estimated token count. Choose a model family (GPT-4, Claude, etc.) for a more accurate estimate. Optionally include special tokens for chat or completion framing.

Your text

Paste or type text below. We’ll show character count, word count, and estimated token count by model family.

Input

Text

Paste or type the text you want to estimate tokens for.

Model family

Token counts vary by model. Choose the family closest to your target API.

Include special tokens (e.g. for chat/completion framing)

Adds a small overhead for message boundaries and system framing.

Estimated token count

0tokens

Character count

Word count

Try DocLD for document AI See features

About this calculator

This calculator is for developers and teams who need to estimate token count from raw text—for prompt design, context window limits, or API cost estimation. Different model families tokenize differently; selecting GPT-4, Claude, Llama, or Mistral gives an estimate aligned with that provider’s typical behavior.

Use the result to stay within model context limits and to approximate cost (tokens × per-token price). For full documents or RAG pipelines, use the PDF Token Size Estimator to estimate tokens from page and word counts.

Approximate tokens per word by model family

Tokenizers vary by model. These are typical approximations for English; other languages may differ.

Model family	Typical tokens per word (approx)	Notes
GPT-4	~1.3	Closer to 1 token per 4 characters for English.
GPT-3.5	~1.3	Similar to GPT-4; subword tokenization.
Claude	~1.2–1.4	Varies by model; often slightly fewer tokens than GPT-4 for same text.
Llama	~1.4–1.6	Often more tokens per word; check specific model.
Mistral	~1.4–1.6	Similar to Llama-style tokenizers.
Default	~1.3	Generic estimate when no model is selected.

Frequently asked questions

Related calculators

PDF Token Size Estimator — Estimate tokens from PDF page and word counts for RAG and chunking.
Document Processing Cost Calculator — Monthly cost for parsing, extraction, OCR, and chat.

About this calculator

Approximate tokens per word by model family

Tokenizers vary by model. These are typical approximations for English; other languages may differ.

Model family	Typical tokens per word (approx)	Notes
GPT-4	~1.3	Closer to 1 token per 4 characters for English.
GPT-3.5	~1.3	Similar to GPT-4; subword tokenization.
Claude	~1.2–1.4	Varies by model; often slightly fewer tokens than GPT-4 for same text.
Llama	~1.4–1.6	Often more tokens per word; check specific model.
Mistral	~1.4–1.6	Similar to Llama-style tokenizers.
Default	~1.3	Generic estimate when no model is selected.