How AI pricing works: tokens, “per million” and subscriptions.
What a token is, why prices are quoted per million, why input and output cost different amounts, and why a chat subscription is a different thing.
This guide explains general ideas. It contains no real prices, because we have not verified any. The rule of thumb for tokens is approximate.
The basics, in plain English
Two different ways to pay
- A subscription is what most people use: a flat monthly fee for a chat app. You don't count anything. What matters is which models the plan includes and what limits it has.
- The API is what developers use to build AI into their own products. They pay for what they use, counted in tokens.
What is a token?
A token is a small chunk of text, often a piece of a word. As a rough rule for English, one token is about three-quarters of a word, so 1,000 words is roughly 1,300 tokens. Other languages and computer code can use more tokens for the same amount of text.
Input and output
- Input tokens are everything you send: your question, any document you paste in, and (in many setups) the earlier messages in the conversation.
- Output tokens are what the AI writes back.
- They are usually priced separately, and output is often priced higher than input. Check each price list, because it varies.
Why “per million”?
One question uses so few tokens that the cost would be a tiny fraction of a penny, so price lists quote the cost of one million tokens instead. To find what a job costs, work out its tokens and scale down.
A worked example with made-up numbers
These numbers are invented to show the arithmetic. They are not any real model's price. Say input costs 2 and output costs 10 per million tokens, in whatever currency the price list uses.
You send a 1,000-word document (about 1,300 input tokens) and get a 300-word summary back (about 400 output tokens).
- Input: 1,300 ÷ 1,000,000 × 2 = 0.0026
- Output: 400 ÷ 1,000,000 × 10 = 0.004
- Total: 0.0066, well under one cent or penny.
Tiny for one job, and it adds up if you do thousands.
Other words you will meet
- Context window: how much text a model can look at in one go.
- Reasoning or “thinking”: some models work through a problem before answering. Settings such as High or Max usually mean more of that. Many providers count the thinking as output, so it can cost more and take longer. Check how each provider bills it.
- Free tier: a free plan, usually with lower limits or different models than a paid one.
What this means for you
A subscription price and a per-million price answer different questions. The first is what you pay to use a chat app. The second is what a developer pays to build on a model. You can't turn one into the other, which is why this site never does.
A lower price per million tokens is not always cheaper in practice. A model that writes longer answers, or thinks for longer, can use more tokens for the same job. And prices are only comparable between models when the currency is the same and you compare input with input and output with output.
This site does not show prices yet. We have no approved source for them, so every result says “{NO_PRICE}” rather than guess.
What to do next
When you look at any price list, check five things: (1) is it a subscription or the API, (2) are input and output priced separately, (3) is it per million tokens, (4) which currency and what date, and (5) which model the plan or price actually covers.
Then find out what the evidence says about which AI suits your job, and compare it with what you would pay.