How much of my Claude plan is left, and how much would I have used without condense? dense usage shows both for your current session, your current week and any limit on a single model, with when each one resets.
dense usageHere's the format. The numbers are just an example.
Claude plan usage
Comparing the same work
Session ───────────────────────────────────────────────────
40% more usage · estimated
Regular Claude █████████░░░░░░░░░░░░░░░░░░░░░░ 28% est.
Dense Claude ██████░░░░░░░░░░░░░░░░░░░░░░░░░ 20%
Resets 2026-09-20 18:30 UTC
This week ─────────────────────────────────────────────────
50% more usage · estimated
Regular Claude ███████████████████░░░░░░░░░░░░ 60% est.
Dense Claude ████████████░░░░░░░░░░░░░░░░░░░ 40%
Resets 2026-09-25 04:59 UTC
█ used · ░ remainingDense Claude is the percentage Anthropic reports for your plan, read with your Claude login. Regular Claude is our estimate of what the same work would have used without condense.
To estimate it, we price the requests that went through condense in that window at list rates, with and without compression, and scale your percentage by the ratio. Output tokens, and requests whose token counts don't match Anthropic's, count the same on both sides. It assumes your plan burns down in line with list prices. Usage outside condense, like claude.ai, gets scaled too, which makes the estimate read high. With no condense requests in a window, you only see the percentage used.
This only works for Claude subscriptions, using the login from Claude Code. API keys have no plan limits to show, and Codex and ChatGPT plans aren't covered. In Claude Code launched through dense, /dense:usage prints the same report. You need dense 0.10.0 or later.