skip to content
a sitting panda, drawn in inkhello panda.

workbench

tools i use, and ideas i’m trying.

using

trying

token usage

i started tracking my token usage in july 2026, to see where it goes and how my model spend changes over time.

  • claude
  • openai
  • kimi
view as table
monthclaudeopenaikimitotal
july 202619,502,076,392062,108,99219,564,185,384
august 202616,963,142,0294,728,503,8051,458,322,64923,149,968,483
september 202623,420,582,61312,633,652,674336,825,62936,391,060,916
october 202613,169,355,39819,107,702,35572,204,65532,349,262,408
november 20262,004,410,0993,670,128,47805,674,538,577

a standing chart, updated as new data comes in. counted from the local logs of claude code, codex and kimi code, cache reads included. claude code clears its logs after 30 days, so months marked estimated are rebuilt from its own daily totals and corrected for the way it counts. chat apps don’t report tokens, so they aren’t included.