Models and tokens, in plain words

Why is a short answer cheap and a long conversation dear? One word: tokens.

5 minLesson 2 of 90 of 3 steps

Before the buttons, two words. Everything you'll see here, from prices to answer quality, comes back to them.

What a model is

A model is a program trained on an enormous amount of text and data that can answer in human language, write code or look at images. Each maker offers several, in tiers, much like cars.

TierCharacterGood for
Small and fastCheap and immediateSimple jobs, summaries, short answers
Mid-rangeBalance of price and qualityMost everyday work
StrongDearer and sometimes slowerHard code, difficult analysis, sensitive writing

A model's full name has two parts: the maker, then the model, such as anthropic/claude-sonnet-4.6 or deepseek/deepseek-v4-flash. The full list with each price is on the models page.

What a token is

Models don't read word by word. Text is broken into small pieces called tokens. A token might be a short word or part of a longer one.

A token is the unit of use: the more a model reads and writes, the more tokens are spent.

One thing that matters for Persian speakers: Persian text usually breaks into more tokens than the same text in English, so your choice of model weighs more on the bill.

Input and output

Every message has two parts, each priced separately:

  • Input: everything you give the model. Your question, an attached file, and the whole conversation so far.
  • Output: the answer the model writes. Output usually costs several times more than input.

Now it's clear why a long conversation gets dearer. A model has no memory, so the whole conversation is sent again with every message. Message twenty means the previous nineteen are read again.

When the subject changes, start a new conversation. Answers get sharper and each message costs less.

Where prices come from

Each model's page lists the price per million input and output tokens, in rials. Rial prices move with the exchange rate; the rate chip at the top of the panel shows the rate each request is billed at.

Don't worry about the price of a single token. Look under the answers instead: in the chat, the real cost of each reply is written right there, and after a few messages you get a feel for it.

Read one model's page0 of 3 steps

  1. Open the models page.
  2. Find a model you've heard of and open its page.
  3. Look at the input and output price and the Capabilities section. Does this model read images?

A quick check

Just for you: your answers aren't sent anywhere.

Why does message twenty of a conversation cost more than its first message?

Which usually costs more per token?

What you learned

  • A model is what answers, and models come in tiers.
  • Tokens are the unit of use; input and output are priced apart.
  • Next: create an account and find your way around the panel.
Creating an account and finding your way aroundNo password and no foreign card. Five minutes from now you have an account and know where everything is.Done, next lesson