Terse logo

Terse

Archives
Log in
Subscribe
May 29, 2026

How I cut my Claude bill by 73%

T
TERSEISSUE 001 · MAY 2026

THE TERSE LETTER · 4 MIN READ

How I cut my Claude
bill by 73%

One config change. Sixty seconds. Most devs are leaving this money on the table.

Most devs using Claude Code are silently burning 70–90% of their context budget on tokens they've already paid for.

Here's the one change that fixes it — and a free tool that shows you exactly when it's working.

01What prompt caching actually does

Claude charges you 10× less for content it's seen before.

Your CLAUDE.md, system prompt, and reference files get cached after turn one. Every turn after that bills at 10% of the normal rate for the cached portion. That sounds incremental. It isn't.

Cost per turn · 10K-token preamble

Without caching$0.150 / turn
 
With caching$0.015 / turn
  
→ At 2 sessions/day, that's roughly $240/year back.
02Why most devs miss it

Your cache hit rate is probably under 30%.

Claude Code caches automatically — but only when content stays stable. New file reads, inline pastes, and shifting context break the cache before it pays off. The fix: stable content first, volatile content last.

✗ BUSTS THE CACHE

Today's date is…
Current task notes
CLAUDE.md
Reference files

✓ CACHE-SAFE ORDER

CLAUDE.md
Reference files
Project conventions
Current task (last)
LIVESee your savings in real time

Terse shows what you're saving, every turn.

Terse floating optimizer bar - captures, optimizes, replaces your prompt with one click

Floating optimizer bar — capture, optimize, replace. Always on top.

03Action items

Three things to do right now.

1

Reorder your CLAUDE.md

Move frequently-changing content — dates, session notes, task lists — to the bottom.

2

Watch your cache hit rate

Terse shows live cache % per turn. Healthy sessions run 60–80%. Under 30% means you're paying full price.

3

Check the cost breakdown

Terse → Statistics shows your savings in dollars per source — not just token counts.

Download Terse — free to start →

macOS · Free forever tier · 1,500 optimizations/week

NEXT WEEK

Git diff compression. Claude Code's git diff output routinely hits 30–60K tokens per turn. I'll show the filter that gets it under 5K with zero information loss.

Until then — stay terse.

— James
Builder of Terse · terseai.org

You're getting this because you subscribed at terseai.org. One email per week. No spam, ever.

Don't miss what's next. Subscribe to Terse:
← Newer Your git diff is costing you 30K tokens
Twitter
terseai.org
Powered by Buttondown, the easiest way to start and grow your newsletter.