Claude usage

Max-plan meter (the number behind /usage), sampled hourly from Mihai's laptop, plus a week-to-week comparison rebuilt from the local transcripts.

Loading…

Weeks

Week to week

API list-price equivalent of every call in the transcript archive, by model. A proxy for the quota, not the meter: Anthropic doesn't publish the weighting. Claude Code prunes old transcripts, so this series starts where the archive does.

Prompts per week, full history

Every prompt typed into Claude Code since the first session, from the prompt log. No tokens in it, but it reaches back where the transcripts don't. Fewer prompts with far more API calls is the delegation era: agents and swarms doing the typing.