The Complete Guide to Tracking Your Claude Usage in 2026

There are four ways to check your Claude usage in 2026: the built-in Settings then Usage page on claude.ai, the /usage slash command inside Claude Code, command-line trackers like ccusage that parse your local Claude Code logs, and Mac menu-bar apps like ClaudeUsageBar and tokn.watch. Each one measures something slightly different and refreshes on its own schedule. Here is what each tool shows, where the numbers come from, and which one fits which workflow.

Update, 2026-09-19: Max plans no longer run a separate weekly limit scoped to Opus. Opus and Sonnet are both current, fully available models, and both draw from the single weekly limit that covers every model. The second weekly meter on Max is scoped to Fable instead: Max includes Fable for up to 50% of the weekly limit, after which Fable keeps running on pay-as-you-go usage credits. Pro has one weekly limit across all models and bills Fable from usage credits from the first message. Checked against a live Claude Max 5x account that used Fable, Opus and Sonnet in the same week: Settings > Usage shows current session, this week, and Fable this week, with no Opus meter.

The four ways to track Claude usage, side by side

Most people learn about a Claude usage tracker the same way: they hit the weekly cap on a Wednesday afternoon, swear loudly, then start Googling. By the next morning they have heard of two or three tools. By the end of the week they have installed one. The problem is that the four main tools are not interchangeable; they measure different things, refresh at different intervals, and are good at different jobs.

Here is the shape of each tool, the question it answers, and the trade-off it makes.

The first three are something you actively check. The fourth is something you glance at while you work. That difference matters more than any of the feature comparisons. If you only realise you should have looked at usage after you have already burnt through it, you do not need a better dashboard; you need an ambient signal.

Below is the detail for each option, with the gotchas that the marketing copy does not mention.

Method 1: The official Settings then Usage page on claude.ai

The fastest path is the one most people miss because it is not promoted: Anthropic ships a built-in usage page inside Claude's web app. To find it, sign in to claude.ai, click your initials in the bottom-left corner, choose Settings, then click Usage in the sidebar.

What you see depends on your plan:

• All plans show the 5-hour rolling session percentage and its reset time, plus the weekly all-models percentage and its reset time.
• Max plans show an additional row for the Fable weekly meter, which sits inside the same weekly window.
• API users see a separate billing-style usage breakdown under the API account section; that is a different surface and reads from a different backend.

This is the source of truth. Every other tool on this list ultimately tries to read or derive the same numbers. Anthropic's usage help-center article describes the page directly and confirms that the percentages are computed against the rolling 5-hour and 7-day windows, not a clock-based reset.

The downside is friction. The Settings page only refreshes when you load it. If you are deep in a Claude Code session and you want to check how much of the weekly cap is left, you have to switch tabs, open the menu, and click through three layers. By the time you are looking at the number, you have lost the thread you were on.

It is also a snapshot, not a stream. Claude updates the underlying numbers a few seconds after each message, but the page does not auto-refresh. If you load it, then send another five messages, the page is now stale until you reload.

Use the Settings page when you want to verify a number you saw somewhere else, or when you have just hit a cap and want to know exactly how long until you can keep working. Do not use it as your primary tracker.

Method 2: The /usage slash command inside Claude Code

If you live in Claude Code, this is the lowest-friction option that exists. Type /usage in any active Claude Code session and Claude prints your remaining session and weekly percentages in the terminal output, no need to leave the editor.

The output looks roughly like this:

Claude usage
Current 5h session: 73% remaining (resets in 2h 18m)
Weekly all-models: 41% remaining (resets in 4d 6h)
Fable this week: 12% remaining (resets in 4d 6h) Example /usage output

Numbers and labels vary between Claude Code releases, but the shape is constant: session line, weekly line, and on Max accounts the Fable weekly line as a third row. The numbers match the Settings page because both are reads against the same Anthropic backend.

What makes /usage better than the Settings page for developers is that it is in the same window where you are working. You do not lose context. The downside is the same: it is on-demand. You only see usage when you actively run the command, and the moment you run it you are also burning a small amount of context window on the response.

One sharp edge worth noting: Claude Code, the Anthropic mobile app, the web app, and the API all share the same usage pool on your account. If you have been chatting on the iOS app while another Claude Code session ran in the background, the /usage number reflects the combined draw. The mental model "I only used Claude Code today, so this is my Claude Code usage" is wrong - it is your account usage.

Method 3: Command-line trackers like ccusage

The third class of tool reads your local Claude Code log files and computes a different angle on the same activity: token consumption over time. The best-known example is ccusage, an open-source Node.js CLI that parses the ~/.claude directory and shows your usage broken down by day, project, and model.

A typical ccusage call looks like this:

• npx ccusage daily - shows today's spend by model
• npx ccusage projects - shows lifetime spend grouped by project
• npx ccusage --since 2026-05-01 - shows a custom date range

ccusage is not a replacement for the Settings page. It does not know about Claude's 5-hour and 7-day caps; it only knows how many tokens passed through your local Claude Code sessions. That is a useful number for cost analysis ("am I making my Pro subscription pay for itself?" or "which client project is consuming the most context?"), but it is the wrong tool for "how close am I to hitting the wall this week". We go through that trade-off in detail in tokn.watch vs ccusage.

Other CLI trackers in this space include CUStats, claude-usage, and cc-stats. They differ in the exact format and visualisation but all share the underlying approach: parse the local logs and aggregate. None of them poll Anthropic's backend; they are pure post-hoc analysis. That means none of them help you decide whether to send the next message - they only tell you what the last hundred messages cost.

Use a CLI tracker when you want to understand where your tokens went: by project, by day, by model. Not when you want to know how much is left.

Method 4: Menu-bar apps that show usage live

The fourth class of tool runs in your macOS menu bar, polls Anthropic's internal usage endpoint every few minutes using your existing browser session, and shows a live remaining percentage and reset countdown without you having to do anything. This is the only category that delivers an ambient signal: you can glance up and see "37% W" without breaking your flow.

The two main entries in 2026 are:

• ClaudeUsageBar - Claude-only. Sits in the menu bar, shows session and weekly percentages, supports both Pro and Max plans. Free during beta.
• tokn.watch - Claude plus ChatGPT in the same menu-bar item. Free. Open source. Reads your existing browser session for both providers. Runs on macOS 13 and up, Apple Silicon and Intel.

The honest comparison between the two: ClaudeUsageBar is Claude-only, which is fine if Claude is your only AI subscription. tokn.watch tracks both providers, which is the differentiator if you also pay for ChatGPT Plus or Pro. Otherwise they read the same backend and converge on the same numbers within a polling cycle.

There are also a few in this category worth knowing about:

• SessionWatcher - cross-platform desktop overlay that surfaces Claude session resets as a small floating widget. Less ambient than a menu bar but useful for second-monitor setups.
• CUStats - mentioned above as a CLI; some forks add a menu-bar mode that displays the local-log summary. It does not poll Anthropic, so the numbers diverge from the Settings page after each new message.

The trade-off menu-bar apps make: they need to authenticate with Anthropic's backend on your behalf. Both ClaudeUsageBar and tokn.watch handle this by reading your existing claude.ai session cookie - the same one your browser already uses - so neither asks for an API key. They are not API-calling tools; they are a different category.

How to choose between them

The right tool depends on three things: where you do your Claude work, whether you also use ChatGPT, and whether you need ambient awareness or just on-demand verification.

Decision shortcuts that hold for most people in 2026:

• You use Claude through the web app, once or twice a week. Use the Settings page. The friction is fine at that volume and you do not need anything else installed.
• You live in Claude Code and just want to glance at the number. Use /usage. It is in the same window, the numbers are accurate, no install needed.
• You hit the weekly cap regularly and want a warning. Install a menu-bar tracker (ClaudeUsageBar or tokn.watch). The ambient signal is the whole point - you will see the percentage falling and pace yourself before it bites.
• You pay for Pro and want to know it pays back. Use ccusage or a similar CLI to compute the per-project cost. The Settings page does not tell you what each project cost; ccusage does.
• You use both Claude and ChatGPT. Pick tokn.watch for the menu-bar slot. It is the only one in this list that tracks both. Run ClaudeUsageBar in parallel if you want a second opinion specifically on Claude.

It is also fine to combine. Many heavy users keep a menu-bar tracker running for the ambient signal and use ccusage every Friday for the per-project breakdown. The tools are not mutually exclusive; they measure different things.

Three things none of these trackers tell you

Even with all four tools running, there are gotchas in Claude's usage model that no current tracker surfaces well. Three of them are worth knowing about:

1. The weekly window is rolling, not Monday-to-Sunday. When you see "weekly all-models: 41% remaining", that 41% is measured against a 7-day window that started on your first message of the current cycle. If you ran out of weekly budget on a Tuesday afternoon, you regain access the following Tuesday afternoon - not midnight on Sunday. This is the single most common misconception about Claude's limits and it bites people who plan around a Sunday-night reset that never comes.

2. Extra usage credits are a separate surface, not a top-up. If you have purchased extra usage credits (the new pay-as-you-go feature on Max), they are spent only after you hit your weekly cap, and they do not show up on the same percentage scale. The Settings page surfaces them as a separate "Extra usage" line. The /usage command may or may not include them depending on Claude Code version. tokn.watch shows the credit balance alongside the percentage when you have extra-usage enabled. The mental model "extra credits add to my weekly cap" is wrong; the model is "extra credits unlock a separate budget that you can spend once the cap kicks in".

3. Max meters Fable separately, on top of the all-models weekly. Max subscribers get two weekly readings, not one: an all-models weekly that covers Sonnet, Haiku and Opus combined, and a Fable meter that is included up to 50% of that weekly limit before Fable moves to pay-as-you-go usage credits. You can be at 80% remaining on the all-models cap and 5% remaining on the Fable meter at the same time. The Settings page shows both rows and /usage shows both rows on Max. Retired: Max plans used to run a separate Opus-only weekly budget in this slot; that meter no longer appears, and Opus draws from the all-models weekly cap like every other model.

For the full mechanics of the underlying caps and what trips each one, see how Claude's usage limits really work in 2026.

What the numbers look like compared to ChatGPT

If you came to this article from a ChatGPT background, the headline difference is that Claude shows you the percentages and ChatGPT does not. There is no equivalent of Settings > Usage on ChatGPT.com. OpenAI does not publish a public usage page on the consumer plans; you find out you have hit the cap by hitting it.

That is the gap menu-bar trackers fill on the ChatGPT side - they read OpenAI's internal endpoints in the same way they read Anthropic's, and surface a remaining percentage that the official UI never shows you. If you use both providers, having one menu-bar item that shows both percentages side by side eliminates a category of "I thought I had budget left" surprises in both directions. tokn.watch does this; ClaudeUsageBar does not.

For the ChatGPT side of the same story, see how ChatGPT's limits really work in 2026.

Frequently asked questions