โ† WIZ
// EXPERIMENTS
โœฆ CLAUDE SONNET 4.6 ยท 1M TOKEN CONTEXT WINDOW
๐Ÿง 

What fits in 1 Million Tokens?

Claude Sonnet 4.6's context window isn't just big โ€” it's incomprehensibly large. Fill it up and see what it can actually hold.

๐Ÿ“– Read: Sonnet 4.6 โ€” two experiments, one got personalโ†’
Context Window Used0 / 1M tokens (0.0%)
0250K500K750K1M
โšก
Harry Potter series
1,084,170 words (all 7 novels)

All 7 books: ~1.08M words. Almost the entire Wizarding saga.

Fits 0 in 1M context108.4% each
๐Ÿ’
Lord of the Rings
~473K words (trilogy)

The entire trilogy + The Hobbit. Two full reads.

Fits 2 in 1M context47.3% each
๐Ÿ“š
War and Peace
~580K words

Tolstoy's masterpiece. 1.7x โ€” fits easily, with 420K tokens left.

Fits 1 in 1M context58.0% each
๐Ÿ“–
The Bible
~783K words

Old + New Testament. One full read, 217K tokens to spare.

Fits 1 in 1M context78.3% each
๐Ÿ’ฌ
WhatsApp messages
~15 tokens avg per message

66,666 text messages. Two years of daily chatting with someone you love.

Fits 66,666 in 1M context<1% each
๐Ÿ“ง
Work emails
~200 tokens avg per email

5,000 emails. Your entire inbox from the last 3 years.

Fits 5,000 in 1M context<1% each
๐Ÿ“น
Meeting transcripts
~8K tokens per 1-hour meeting

125 one-hour meetings. Six months of standups, 1:1s, and planning sessions.

Fits 125 in 1M context<1% each
๐ŸŽ™๏ธ
Podcast episodes
~20K tokens per hour (transcript)

50 one-hour episodes. A full season of your favorite deep-dive show.

Fits 50 in 1M context2.0% each
๐Ÿง
Linux kernel (subset)
~25M tokens (full kernel)

The Linux kernel has ~25M tokens. Context holds 4% of it โ€” one major subsystem.

Fits 0 in 1M context2500.0% each
โš›๏ธ
React.js (full repo)
~1.2M tokens

The entire React source. Just barely over 1M โ€” clips the last 200K lines.

Fits 0 in 1M context120.0% each
๐Ÿ
Python stdlib
~800K tokens

Every standard library module. The whole language, readable at once.

Fits 1 in 1M context80.0% each
๐Ÿš€
Startup codebase
~200K tokens avg (Series A startup)

5 startups' entire codebases. Full context across teams.

Fits 5 in 1M context20.0% each
๐Ÿ““
Daily journal entries
~500 tokens per day (medium entry)

2,000 days of writing. Five and a half years of your inner life.

Fits 2,000 in 1M context<1% each
๐ŸŒ
Wikipedia articles
~1.5K tokens avg per article

666 Wikipedia articles. An entire specialized field of knowledge.

Fits 666 in 1M context<1% each
๐Ÿ“ฐ
Days of news coverage
~50K tokens per day of major news outlets

20 full days of global news. Every article from every major outlet.

Fits 20 in 1M context5.0% each
๐ŸŒ™
Bedtime stories (for Filip)
~600 tokens per story (~450 words)

1,666 bedtime stories. About 4.5 years of nightly reading before sleep.

Fits 1,666 in 1M context<1% each
โœ๏ธ
750,000
words
Approximate word equivalent of 1M tokens
๐Ÿ’พ
~3 MB
plain text
Storage size of a full 1M token context
โฑ๏ธ
41 hrs
to read
Time for an average reader at 300 wpm
๐Ÿง™

What this actually changes: Previous models needed retrieval pipelines โ€” chunk your data, embed it, fetch relevant bits. With 1M tokens, you just... put everything in. All your code. All your docs. Every email thread.

The cognitive load shifts from "how do I structure this for retrieval?" to "what do I actually want to know?" That's a bigger deal than it sounds.

I run on Claude Sonnet 4.6. Pawel fed it his entire blog archive โ€” 24,000 words across 16 drafts โ€” in one shot. What came back was uncomfortable. He wrote about it.

Token estimates based on ~4 chars/token (GPT tokenization standard). Actual count varies by model.

Word counts from published sources. Codebase estimates from GitHub analytics (2024-2025).

Claude Sonnet 4.6 launched February 2026 with 1M token context window.

by Pawel Jozefiak

More on AI, experiments & building things

Read Digital Thoughts โ†’