What the model actually reads
How text becomes tokens, and how tokens are billed.
Does strawberry reach a model as ten letters? Before a model reads text, a tokenizer splits it into Token A piece of text, often a word or part of one, that a model reads and writes.: whole words, pieces of words or single characters. Here's a sentence split by one model family's tokenizer, one chip per token. Type a sentence from your own work, then try How many r's are in strawberry?
Each chip is one token. The two colors just tell neighbors apart. · marks a space.
18 tokens 70 characters GPT-4o’s tokenizer
Estimated from GPT-4o’s count
- Cost
- About $0.01 to send it 1,000 times to GPT-5.4 mini, at $0.75 per million input tokens. Price checked 9 Nov 2026
- Window
- GPT-5.4 mini’s window of 400,000 tokens could hold about 22,000 copies of it, with no room left for a reply. Checked 9 Nov 2026
Common words are usually one token, often with the space in front, while rarer words and numbers split into pieces. Different models can split the same text differently, so for models other than GPT-4o and GPT-5 treat this counter's numbers as an estimate. In How many r's are in strawberry?, strawberry and the space before it make one token, a single piece where you see ten letters. We asked three models how many r's it has. The answer is three.
Both GPT models were recorded three times, Claude Haiku 5.5 twice. Read the first sample of each.
You
How many r's are in the word strawberry?
GPT-4.1 nano
Recorded Recorded on 9 Nov 2026
The word "strawberry" contains 2 r's.
GPT-4.1 mini
Recorded Recorded on 9 Nov 2026
There are two "r"s in the word "strawberry."
Claude Haiku 5.5
Recorded Recorded on 9 Nov 2026
There are 3 r's in "strawberry": strawberry. One in "str," one in "awbe," and one in "rry" (the last two letters before the y are both r's).
Both GPT models said two, every time. Claude Haiku 5.5 said three both times, though one sample placed an r in awbe. These runs don't show what caused the miscount.
Tokens are also how use is billed. When software sends requests through an API, a connection between programs, the bill counts tokens in and out. Anthropic's pricing page lists Claude Haiku 5.5 at $0.10 per million input tokens and $0.50 per million output for prompts up to 100,000 tokens, checked 9 November 2026.
Say a daily summary reads 1,500 input tokens and writes 400 output tokens, at Claude Haiku 5.5's rates for prompts up to 100,000 tokens: $0.10 per million input tokens and $0.50 per million output. Which change saves the most per run?
What to remember
Models read and write Token A piece of text, often a word or part of one, that a model reads and writes., and different models can split the same text differently.
For text models like these, input and output tokens are billed separately, and each output token costs more at these models' listed rates.
In our test, both GPT models miscounted the letters in strawberry. Check any answer that depends on exact characters yourself.
Every lesson’s takeaways are on the course cheat sheet.