Skip to content

Working with clients? Stay ahead of what they ask for.

No. 012 · 7 November 2026

Where do AI coding tokens go?

How AI coding agents’ tokens split between re-reading context and writing new code, and how that shows up on the bill.

Infographic No. 012: Where do AI coding tokens go? A grid of 1,000 squares, each 0.1% of tokens: 96% re-reading, 3.6% fresh input and cache writes together, 0.4% written. They read 240 tokens for every 1 they write. A written token costs 50 times a read one, yet re-reading and caching are 83% of the bill. What your agent re-reads is what you pay for.

AI coding agents spend 96% of tokens re-reading and 0.4% writing code: 240 read for every 1 written. Re-reading and caching are 83% of the bill.

The numbers behind it

AI coding tokens and cost, H1 2026
MeasureFigure
Tokens spent re-reading (cache reads)96%
Tokens for fresh input and cache writes3.6%
Tokens written as output0.4%
Bill for cache reads and writesabout 83%
Bill for output16%
Median user’s monthly cost, August 2025about $5
Median user’s monthly cost, May 2026$81

Source and method

Jellyfish data for the first half of 2026, rounded as published. “240 for every 1” and “50 times” are our arithmetic; the 3.6% is what remains after re-reading and output, as Jellyfish does not split it.

What it means

  • Context drives cost: what your agent re-reads is what you pay for.
  • Keeping an agent’s context small and focused is the direct way to cut the bill.
  • The median user’s monthly cost rose from about $5 to $81 in nine months.
Was this useful?

Embed this infographic

Paste this into your page or post. It shows the infographic and credits it with a link back here.

Part of Visual Explainers, the Automation Switch collection of infographics. Free to share and embed with a credit link back to this page.