Language Models

Token

The bite-sized pieces of text an AI reads and writes: often a word, sometimes part of a word.

In everyday terms

"Unbelievable" might be split into "un", "believ", "able". Roughly, 1 token ≈ ¾ of an English word. Pricing and limits are counted in tokens.

For professionals

Sub-word units from a learned vocabulary (e.g. BPE). Each token maps to an ID and then an embedding vector.

Think of it like…

Lego bricks of language. The AI builds every sentence brick by brick.

You've already seen it

"Max 4,000 tokens", API prices per million tokens, odd spelling mistakes on rare words.

Myth vs reality

Myth: AI reads letters like we do.

Reality: It sees tokens. That's why it can struggle to count letters in a word.

Quick check

Roughly how many English words is 1,000 tokens?

Show answer

About 750: 1 token ≈ 0.75 words.

Builds on

Large Language Model (LLM)

Related

Context Window · Large Language Model (LLM) · Embedding

🔎esc