Tokens, limits and what you pay for
Key takeaway: You are billed for text in and text out, counted in tokens — roughly three quarters of a word each.
What a token is
Models do not read letters or words; they read tokens, which are common chunks of text. In English a token averages about four characters, so 1,000 tokens is roughly 750 words — a page and a half.
Both directions count. A long document you paste in is input tokens, and the reply is output tokens. Output is usually priced two to five times higher than input, which is why asking for a summary is cheap and asking for a full rewrite is not.
The three numbers on every pricing page
- Price per million input tokens, and separately per million output tokens.
- Context window: the largest amount of text the model can hold in one conversation.
- Rate limits: requests or tokens per minute, which cap how fast you can work, not how much you pay.
Seats versus usage
Consumer plans (about $20 a month per person) bundle generous usage into a flat seat price and are the cheaper option for individuals. API pricing charges per token and only wins when usage is either tiny or automated at scale.
A quick sanity check: 100 requests a day, each with a 2,000-token prompt and a 500-token answer, is roughly 7.5 million tokens a month. At typical mid-tier prices that is a few dollars — far less than a seat. Automation is cheap; humans reading and writing all day are not.