Data

My AI usage, in the open

Token volume, harness breakdown, model evolution, and cache efficiency — snapshotted fromccusageat the end of each month. I'm on a flat Team subscription, so the API-equivalent dollar values are hypothetical — but the token counts and patterns are real.

Last updated: May 2026

Total tokens

7.1B

across 4 months

API-equivalent value

$4167

at API rates (subscription pays flat)

Cache hit rate

98%

reads vs writes — higher is better

Harnesses used

4

Claude Code, Codex, OpenCode, Hermes

Monthly token volume by harness

Harness distribution

Token volume by model provider

Model evolution — which models I actually used each month

Same vendor, different model each month shows AI tooling churn. Colors group by provider family; shades distinguish models within the same vendor.

Cache efficiency — writes vs reads

Cache reads are ~10× cheaper than writes. A healthy session has a small write spike at start, then overwhelmingly reads.

0:000:00