Token pricing
In short
Token pricing is how AI companies charge developers for using their models: you pay for each token, a small chunk of text, with one rate for the text you send in and a higher rate for the text the model writes back.
- Also called
- Per-token pricing, API pricing, Pay per token
- Topic
- Using AI tools
- Level
- Beginner
A token is a piece of text a model reads or writes. OpenAI's rule of thumb is that one token is about four characters of English, or three-quarters of a word, so 100 tokens is roughly 75 words1. The exact count varies by model and language1.
Prices are quoted per million tokens, and output costs more than input. On OpenAI's price list in October 2026, GPT-6 Astra costs $10 per million input tokens and $50 per million output tokens, while the smaller GPT-6 Luna costs $0.10 and $0.502. Anthropic lists Claude Opus 5.5 at $4 for input and $20 for output3.
Discounts can change the bill a lot. Both companies charge half price for batch jobs that do not need an instant answer23. Prompt caching makes repeated text cheaper: on Anthropic's API, re-reading cached text costs 10% of the normal input price on most models3.
Some tokens are invisible. Reasoning models "think" in tokens you never see, and OpenAI bills those as output tokens4. That is why a short answer from a reasoning model can cost more than expected.
If you only use a chat app on a monthly plan, you will rarely see these prices. They matter when you, or a business, build AI into software through an API.
Example
Summarising a 3,000-word report means sending about 4,000 tokens and, for a 400-word summary, getting about 530 back1. On GPT-6 Luna that costs well under a tenth of a penny2. The same job on Claude Opus 5.5 costs about 2.7 US cents3.
Why it matters
It decides what AI features cost to run. Long documents, long chats and reasoning models all use more tokens, so the same task can cost a fraction of a penny or many times more depending on the model.
Questions people ask
How many words is a million tokens?
Roughly 750,000 words of English, using OpenAI's rule of thumb that 100 tokens is about 75 words1. Other languages and code can count differently, and each model's tokeniser splits text in its own way1.