AI Token Counter & Cost Estimator

What is an LLM token?

A token is the basic unit large language models read and produce. Tokens are not characters or words — they're sub-word pieces produced by a tokenizer. Common English words like "hello" are usually one token; rarer words split into multiple. As a rough rule:

• 1 token ≈ 4 characters in English
• 1 token ≈ ¾ of a word
• 100 tokens ≈ 75 words

Code and non-English text typically use more tokens per character.

Why token count matters

• Pricing. API providers charge per token — both for what you send (input) and what you get back (output). Output tokens often cost 3–5× more than input tokens.
• Context limits. Each model has a maximum context window (e.g., 128K, 200K, 1M tokens). Long prompts plus long histories can exceed this.
• Latency. More tokens = more time to generate.

Tokenizers vary by model

Different model families use different tokenizers. OpenAI uses tiktoken (cl100k_base for GPT-4 family); Anthropic uses a similar BPE tokenizer; Google's Gemini uses SentencePiece. Counts can differ by 5–15% across families. Our estimate uses an average heuristic — for exact billing-grade counts, use the official tokenizer for your model.

Frequently Asked Questions

How accurate is this estimate?

Within ~10–15% for typical English prose. Code, JSON, non-Latin scripts and unusual formatting can skew it more. Treat the result as a budgeting aid, not an exact bill.

Why are output tokens more expensive?

Generating tokens requires running the full forward pass through the model for each token, while input tokens can be processed in parallel. This makes output 3–5× more compute-intensive.

How can I reduce my AI bill?

Use a smaller model when possible (Haiku, Mini, Flash variants are 10–50× cheaper). Trim your prompt — remove unnecessary context, examples and instructions. Cap max_tokens on outputs. Use prompt caching where supported (Anthropic and OpenAI both offer it).