What Does 1 Million Tokens Look Like?
One million tokens is the standard pricing unit for AI APIs, but the number is too large to visualize on its own. This guide translates it into things you can picture — books you have read, code you have written, conversations you have had.
At a glance
One million tokens is approximately:
750,000
English words
1,500
pages of text
8
average novels
100,000
lines of code
83
hours of speech
1,000
images analyzed
In books you know
The easiest way to feel the scale is through books. One million tokens is enough to hold:
- 8 average novels — a typical novel is about 90,000 words (120,000 tokens)
- 69% of Harry Potter — the full series is 1.08 million words (1.45M tokens), so one million tokens gets you through Order of the Phoenix
- 1.3 copies of War and Peace — Tolstoy's epic is ~587,000 words (780,000 tokens)
- The full Lord of the Rings with room for a half-reread — the trilogy is 481,000 words (641,000 tokens)
- 96% of the King James Bible — at 783,000 words (1.04M tokens), you almost fit the whole thing
In words and pages
750,000 words is a genuinely enormous amount of text. For perspective:
- The average person reads about 250 words per minute — reading 750,000 words would take 50 hours of non-stop reading
- Printed single-spaced, it would be 1,500 pages — a stack about 15 cm (6 inches) tall
- It is roughly 1,000 Wikipedia articles at an average of 725 words each
- Or about 10,000 work emails at 75 words per email
In source code
Code tokenizes differently from prose — syntax characters, indentation, and variable names all inflate the token count. On average, one line of code is about 10 tokens, so one million tokens is roughly 100,000 lines of source code.
For reference, that is:
- Nearly twice the original Doom engine (57,000 lines)
- About the size of a mid-sized production web application
- Roughly 2,500 average source files at 40 lines each
This means frontier models with 1M+ context windows can theoretically hold an entire mid-sized codebase in a single prompt — though in practice, you would use retrieval to focus on the relevant files.
In speech and audio
People speak at roughly 150 words per minute. One million tokens of transcribed speech equals about 83 hours of talking — three and a half days straight. That translates to:
- 110 podcast episodes at 45 minutes each — over two years of a weekly show
- 2,500 songs' worth of lyrics at ~300 words per song — over six days of continuous music
What it costs
The same million tokens costs anywhere from $0.50 to $30 for input, depending on the model. That is a 60x price spread for the exact same amount of text.
How it relates to context windows
A model's context window is the maximum amount of text it can process in a single API call. Most frontier models in 2026 support 1M+ tokens:
- GPT-5.6 Sol: 1.05M context window
- GPT-5.5 Pro: 400K context window
- Claude Fable 5: 1M context window
- GPT-5.6 Terra: 1.05M context window
- GPT-5.5: 400K context window
A million-token context window means you can send an entire codebase, a full book, or months of conversation history in a single request. Budget models typically have smaller windows (128K-200K tokens) but cost far less per token.