This dashboard is currently showing sample data, not real usage. See the About page for details.
Usage data as of Sep 15, 2026 · Pricing last confirmed Sep 15, 2026 · Reviewed weekly

About TokenWatch

TokenWatch analyzes real Claude usage (from OpenClaw automation and Claude Code sessions, plus manually logged Claude.ai usage) and compares it against Claude Pro ($20/mo), Claude Max ($100/mo), and pay-per-token API pricing, to answer a simple question: which one is actually worth it, given real usage, not a guess.

This dashboard is currently showing sample data. The historical-log ingestion that replaces it with real numbers is separate, ongoing work. Every number here is illustrative, not a real claim about anyone's actual spend.

How the numbers are calculated

Every logged interaction records input, output, cache-read, and cache-write token counts, tagged by model. Those counts are priced against pricing_reference, a table of currently published per-model rates, to produce an "equivalent API cost", which is what's compared against Pro and Max's flat monthly rates.

That comparison only means anything over a matching window, so every figure on this dashboard covers the current calendar month to date: the 1st of the month through today. Early in a month the total is a partial month and will be lower than the eventual full-month figure.

Tokens that ran on local models are excluded from the cost totals. They run on hardware that is already paid for, so their per-token cost is zero, not unknown. They still appear in the usage breakdowns and in the local-vs-cloud split on the overview.

If no usage has been recorded for the current month, the affected pages say so outright rather than showing a $0 total and a recommendation drawn from it.

Sources