Claude usage & cost history

Loading…

Claude Code — daily token burn by model

Local Claude Code session transcripts only — subagent (Task-tool) token usage is not captured in these logs and is missing from this chart. Window is capped by local retention (~40 days); older transcripts are already pruned from disk.

Anthropic API org — real $ spent (daily)

Ground-truth dollars from the Anthropic Admin Cost API for the ToolForge API org (pipeline/batch jobs) — the account that ran the March classification job. Not Claude Code.

Claude Max plan — real invoices (monthly)

Manually captured from claude.ai billing (no public API for individual Max-plan invoices). Powers Claude Code sessions.

Claude Code — tokens by model (all-time on disk)

Current cycle — Max plan limits

Live snapshot captured 2026-07-25. No historical time series exists for this — Anthropic doesn't expose it, and it resets weekly/monthly. This is a point-in-time reading, not tracked continuously (yet).

By project — Claude Code sessions, active hours, tokens, modeled cost

Project	Sessions	Active hrs	Tokens	Modeled $	Last active

"Modeled $" = session tokens × live API list pricing per model, as if this Max-plan usage had been billed pay-as-you-go. The Max plan is a flat subscription (see invoices above), so this is a value/comparison figure, not money actually spent.