A million tokens: what changes when AI can read an entire book at once
Over the past few weeks, several major labs have unveiled language models with a context window of a million tokens. Until recently that was a research-lab figure; today it is becoming standard for top-tier models.
What context is
Context is the amount of text a model can keep in mind while it answers. A million tokens is roughly 700,000 words: about the length of an entire fantasy saga, or the complete code of a mid-sized application.
With small contexts, you had to split documents up, summarize them piece by piece and hope nothing important got lost. With a context this large, the model sees everything at once: the contradictions between chapter 2 and chapter 30, the function written three years ago that breaks the new one, the clause buried on page 140 of a contract.
Where the difference shows
Software development: the model can reason about an entire project, not just the file that is open.
Writing: review of a complete manuscript, checking consistency of characters, timelines and places.
Office work: analysis of document archives, meeting minutes and spreadsheets without having to pick them out by hand.
Creative work: screenplays, game worlds and series bibles that stay consistent from start to finish.
The limits to know about
More context does not automatically mean better answers. Reading a million tokens costs more and takes longer, and models still tend to give more weight to the beginning and end of a text than to the middle. That is why labs are cutting the price of repeatedly reading the same documents, known as caching: anyone who works on the same project for days pays far less the second time.
What it means for creators
The real news is not the number but the way of working: you move from single questions to whole projects handed to an assistant that remembers everything. Those who organize their materials well, with tidy files, clear notes and dated versions, will benefit the most.