Personal telemetry · May 2026 — Jul 2026

A ledger of tokens

Every LLM call I made — split between work and personal use, by model and by harness. Updated weekly.

67% of my LLM usage is work — and 61% of all tokens are served from cache, not recomputed. Most of what I ask has already been asked before.

WorkPersonal
Total tokens
2.7B
894M / month avg
Work share
67%
1.8B work · 881M personal
Cache efficiency
61%
served from cache, not recomputed
Peak month
1.3B
Jun 2026
Busiest day
171M
Jul 1, 2026
Yearly trajectory
annual totals
2026baseline
2.7B
Work 1.8BPersonal 881M
Monthly volume
tokens / month
0500M1B1.5B
Daily rhythm — last 12 months
work vs personal, own scale each
Work
AugSepOctNovDecJanFebMarAprMayJunJul
lessmore
Personal
AugSepOctNovDecJanFebMarAprMayJunJul
lessmore
Efficiency over time
2026
Cache-read share61.4%baseline
Output : input ratio0.015baseline
Tokens / session4.7M

Lower output:input and rising cache-read share both indicate more reuse and less re-explaining context.

By model
claude-sonnet-4-6830M
claude-opus-4-7788M
claude-opus-4-8532M
claude-sonnet-5429M
claude-fable-5309M
claude-haiku-4-5259M
<synthetic>24M
gpt-5.519M
By harness
Copilot CLI2B
Claude Code1.2B

Bars share one scale; the split shows work vs personal within each row.

Token composition — input · output · cache read · cache write
Work
Input 899MOutput 6.8MCache read 833MCache write 64M
Personal
Input 1.8MOutput 6.7MCache read 815MCache write 57M
Data refreshed weekly from api.bipin.io. Tokens counted across every harness and model I use — work and personal. Last updated July 27, 2026.