Verified against Anthropic documentation · Updated October 3, 2026

How to check your Claude usage

To check your Claude usage in the web or desktop app, click your profile avatar and open Settings › Usage. You'll see your 5-hour session bar, your weekly limits, and when each one resets. In Claude Code, run /usage in the terminal for session tokens, an approximate cost, and where you stand against your plan limits.

What to check and where

One shared allowance: on Pro, Max, Team, and seat-based Enterprise plans, the Claude apps and Claude Code draw from the same 5-hour and weekly limits.
Claude app: Settings › Usage is the quickest way to see both limits and when they reset.
Claude Code: /usage adds what the app doesn't show: tokens, cost, and which skills, subagents, plugins, and MCP servers used them.
While you code: a statusline can keep 5-hour and 7-day percentages in your terminal footer (Pro and Max, after the first response).
API keys: pay-as-you-go API usage shows up only in the Anthropic Console, and subscription usage never does.

Claude web & desktop

Profile avatar → Settings › Usage

Live bars for your 5-hour session and weekly limit, with reset times, plus an Opus bar if your account shows one. No token counts and no history.

Steps and what you'll see ↓

Claude Code (terminal)

/usage

Session token counts, an approximate cost, your plan usage bars, and a Day/Week breakdown from this machine's sessions. For a reading that stays on screen, add a statusline.

What /usage shows and misses ↓

Which one should you check right now?

Chatting in the Claude app“I want to know if I'm close to running out.”

Settings › Usage

Working in Claude Code“I want this session's tokens and cost, and what used them.”

Run /usage

Coding for hours“I want my limits and reset time in the terminal footer while I work.”

Set up a statusline →

Paying per token with an API key“I need monthly spend and usage charts.”

Anthropic Console

Want it on screen all the time“I want menu-bar percentages and pacing warnings across all my Claude apps.”

Install our tracker →

What each method shows, side by side

✓ shows it, ~ shows it only in part or on some accounts, ✕ doesn't show it.

Method	Best for	5-hour session	Weekly limit	Opus limit	Reset time	Tokens & cost	History & attribution	Pacing	When you see it
Claude web & desktop Settings › Usage	A quick look at your plan limits	Yes: Progress bar and time left	Yes: All-models progress bar	Partly: Separate Opus bar on accounts that show one	Yes: Reset day and time	No: No tokens or cost; usage credits balance only	No: Current windows only. No history or per-project split.	No: No burn rate or forecast	When you open Settings
Claude Code /usage command	Session tokens, cost, and what used them	Yes: Plan usage bars	Yes: Plan usage bars	No: Aggregate bars only; no separate Opus bar	Yes: 5-hour and weekly reset dates/times when subscription usage data is available	Yes: Session tokens and estimated cost	Partly: Day/Week split by skills, subagents, plugins, and MCP servers, from this machine only	No: No burn rate or forecast	When you run it in the terminal
Claude Code statusline statusLine footer	Keeping limits in view while you code	Partly: five_hour % (Pro/Max, after the first response)	Partly: seven_day % (Pro/Max, after the first response)	No: No Opus field	Partly: resets_at, formatted as local time (Pro/Max)	Yes: Context tokens and session cost	No: Nothing kept across sessions	No: Context % only; no burn rate	Always, in the terminal footer
Anthropic Console console.anthropic.com	API token usage and billing	No: None for API keys	No: None for API keys	No: None; usage is priced per token by model	No: No plan reset; monthly billing and per-minute rate limits	Yes: Exact tokens, cache hits, and spend	Yes: Usage and cost by day, month, and workspace	Partly: Spend limits and alerts; no burn rate or forecast	When you open the web dashboard
Claude Usage tracker (ours) Menu bar / localhost	Limits and burn rate always on screen	Yes: Live % and countdown	Yes: Live % and reset time	Partly: Model-specific windows when Anthropic returns them	Yes: Countdown and local time	No: Percentages only; no tokens or cost	No: In memory only. No history or project split.	Yes: Burn rate in %/day and %/hour	Always: menu bar, 127.0.0.1 dashboard, CLI

For how the limits themselves work, see the 5-hour session limit, weekly limits, and plan limits across Free, Pro, and Max.

How to check each one, and what it won't tell you

1. Claude web and desktop app

1Open claude.ai or the Claude desktop app.
2Click your profile avatar or initials in the bottom-left corner of the sidebar.
3Choose Settings, then open the Usage tab.

What you'll see

5-hour session bar: the percentage used and the time left in the rolling session window.
Weekly all-models limit: how much of the week you've used, and your account's reset day and time.
Opus bar (some accounts): some accounts show a separate Opus weekly meter. Treat it as specific to your account, not an extra allowance everyone gets.
Usage credits: set up or check pay-as-you-go credits that let you keep chatting after a limit runs out. See the usage credits guide.

If you run out mid-conversation, Claude shows a notice in the chat with the exact time your limit resets. The reset timing guide explains how that time is set.

What it can't show

Token counts. Subscription usage is metered dynamically, so you get percentages instead.
History. Only the current 5-hour and weekly windows.
Which project used what. It's account-wide; nothing breaks usage down by project, document, or chat.
Pacing. No burn rate and no forecast of when you'll hit the limit.
API key usage. That's only in the Anthropic Console.

2. Claude Code: the /usage command

Run /usage in any Claude Code session. /cost and /stats are aliases for the same command.

# Inside a Claude Code session:

> /usage

What you'll see

Current session: token counts (input, output, cache reads and writes) and an estimated cost at list rates. /clear resets these.
Plan usage bars: on Pro, Max, Team, and Enterprise, where you stand against your included allowance.
Day and Week breakdown: the last 24 hours and 7 days, split by skills, subagents, plugins, and individual MCP servers.
Behavior flags: notes such as long context, cache misses, and scheduled loops, when they apply.

What it can't show

Other machines. The Day and Week breakdown comes from session history stored on this machine. Usage on other computers or in claude.ai isn't included.
Long-term history. You get the last 24 hours and 7 days, not a permanent account-wide record. The Day/Week breakdown survives /clear; the session totals don't.
A separate Opus limit. The plan bars are aggregate.
A forecast. It shows what used your allowance, not whether the rest will last until the reset.

Why Claude Code's cost total can be higher than your transcript

Claude Code can count model work that never shows up as a request in the main local transcript (the JSONL files under ~/.claude/projects/). In our October 3, 2026 test, a cost rebuilt from the transcript came in 5.65% below Claude Code's own model-usage total. The gap was mostly Haiku 4.5 usage during WebSearch and extra Opus requests after each turn. This is about token and cost accounting; the test didn't show any effect on your 5-hour or weekly percentages. Read the full study.

What the test measured

Claude Code 2.1.289 in an Anthropic-hosted cloud session, Claude Opus 5.5 on high effort, prompt suggestions on. We split the activity by turn and compared Claude Code's per-model totals with the model requests in the local JSONL transcript.

Observed work	Counted by Claude Code?	A request in the main JSONL?
Main Opus conversation requests	Yes	Yes
WebSearch-associated Haiku usage	Yes	No
WebFetch Haiku request	Yes	No
Extra post-turn Opus request	Yes	No

Turn 1, no web calls: Claude Code's totals and the transcript matched exactly.
Turn 2, WebFetch: one fetch produced a Haiku request of 226 input and 81 output tokens; a ~67 KB documentation fetch returned its content with no Haiku request.
Turn 3, WebSearch: Claude Code attributed 24,725 input and 848 output tokens to claude-haiku-4-5-20251001 across two searches ($0.048965 at list price). The transcript's web tool records logged the calls but carried no Haiku token accounting. An earlier cloud session showed 21,633 Haiku input and 780 output tokens across two searches, so all four WebSearch calls we observed came with Haiku 4.5 usage.
After each turn: Claude Code's totals included one extra Opus request missing from the transcript: 506 new input tokens, a cache read of roughly 149k to 160k tokens, 2 to 93 output tokens, about $0.034 to $0.037 at list price. We didn't establish what these requests were for.
Net over three turns: about $1.987 rebuilt from the transcript against about $2.106 in Claude Code's own total, a $0.118986 (5.65%) gap.
Context counter: get_session reported 0 context through the first turn; after that it equalled the previous turn's final request context in 3 of 3 observations.

Scope: one three-turn session, plus WebSearch behavior from one earlier cloud session. Four WebSearch calls across two sessions show what happened in those runs, not how WebSearch always behaves or a fixed per-search token cost across Claude Code versions and environments.

The Claude Code usage guide goes further into terminal limits, context compaction, and subscription versus API billing.

3. Claude Code statusline

If you'd rather not keep typing /usage, set up a statusline. Claude Code runs your script locally as assistant messages arrive and passes it session JSON on standard input. The script uses no API tokens and makes no network calls. The footer can show:

Context fill: how full the context window is (0–100%), so you know when to run /compact.
5-hour and 7-day limits: rate_limits.five_hour and rate_limits.seven_day carry used_percentage and an epoch resets_at timestamp you can format as local time.
The basics, on any account: model, git branch, context fill, and session cost still show when the rate-limit fields are missing.

What it can't show

Limits on Team, Enterprise, or API keys. The rate-limit fields appear only for Claude.ai Pro and Max subscribers, or behind applicable Claude apps gateways.
Anything before the first response. The fields stay absent until the first assistant response completes, and either window can be missing on its own.
History. It's visible only while Claude Code is running in your terminal, and nothing carries across sessions.
An Opus limit. You get the five-hour and seven-day meters only.

The Claude Code statusline guide has a tested Bash script with jq parsing, local reset times, and ANSI threshold colors.

4. Anthropic Console (API keys)

If you call the API with your own keys, or run Claude Code with ANTHROPIC_API_KEY, check usage at console.anthropic.com › Usage:

Token charts: input, output, cache-read, and cache-write tokens by day and month.
Cost and billing: month-to-date spend, invoice history, and prepaid balance.
Rate limits: your tier's requests per minute (RPM) and tokens per minute (TPM).
Workspace attribution: filter usage by API key, workspace, or team member.

The Console doesn't show Pro, Max, or Team subscription usage at all. API requests have no 5-hour or weekly limit; RPM and TPM limits apply instead. Full organization billing and workspace reports need an owner or billing role.

5. Our local tracker (menu bar and localhost dashboard)

Settings › Usage and /usage only show your usage when you stop and look. Our free tracker keeps it on screen: in the macOS menu bar, on a localhost dashboard (127.0.0.1) that refreshes itself, or in the terminal.

Menu bar: live 5-hour and weekly percentages with reset countdowns.
Pacing: a sustainable rate in % per day and % per hour, to help you avoid running out early.
Model-specific windows: account-specific Sonnet or Opus weekly windows when Anthropic returns them.
Privacy: it reuses your local Claude Code sign-in and sends no credentials or usage data to claudeusage.dev.

What it can't do

Keep history. It runs in memory, listens only on 127.0.0.1, and doesn't store a multi-week log.
Split usage by project. It reads your plan quota from Anthropic's internal OAuth endpoint, so it can't attribute usage to folders or git repos.
Count tokens. You get quota percentages and reset countdowns, not token numbers.
Track API keys. It covers subscription usage only, not ANTHROPIC_API_KEY usage or Console balances.
Run without a local install. It needs macOS or Linux with Python 3.9+ and Claude Code on your PATH.

Install the tracker (about 30 seconds in the terminal).

Plan limits, Claude Code costs, and API billing are different numbers

Each one is measured differently and checked in a different place.

Subscription

Your plan allowance

Pro, Max, Team, and seat-based Enterprise plans.
Rolling 5-hour session and weekly limits, shown as percentages.
Check it in Settings › Usage, the /usage plan bars, or a statusline.
Never appears in the Anthropic Console.

Claude Code

Session tokens and cost

The /usage session block and the statusline cost field.
Token counts and an estimated cost at list prices.
Separate from your plan percentage.
Can include model work your local transcript doesn't record (see above).

API

API key billing

Developer API keys (ANTHROPIC_API_KEY), including Claude Code signed in with one.
Billed pay-as-you-go per token at list rates.
Check it in the Console's Usage and Billing pages.
Doesn't touch your 5-hour or weekly limits.

Frequently asked questions

Where is Settings › Usage in the Claude app?

Click your profile avatar or initials in the bottom-left corner, choose Settings, then open the Usage tab. It shows your 5-hour session bar with the time left, your weekly limits, and when each one resets.

Does Claude Code usage count toward the same limits as Claude.ai?

Yes, on Pro, Max, Team, and seat-based Enterprise plans with included usage limits. Claude Code and the Claude apps draw from the same 5-hour session and weekly limits. If you sign Claude Code in with an API key (ANTHROPIC_API_KEY), that usage is billed per token through the Anthropic Console instead.

Why doesn't Claude show exact token counts for my subscription?

Subscriptions don't come with a fixed token number. Anthropic meters them based on conversation length, file attachments, model choice, thinking effort, and current demand, so Settings › Usage shows progress bars and countdowns instead. Per-token billing only applies to API usage, which you check in the Anthropic Console.

Why doesn't my Claude Pro or Max usage appear in the Anthropic Console?

The Console only covers API keys and per-token billing. Subscription usage is tracked separately and never appears there. Check it in Settings › Usage or with /usage in Claude Code.

Can I see historical usage charts for my Claude subscription?

Not a full history. Settings › Usage shows only your current 5-hour and weekly windows. In Claude Code, /usage breaks the last day and week down by skills, subagents, plugins, and MCP servers, but only from sessions on the current machine. Neither gives you a multi-week record across devices. API accounts get full token and cost history by day, month, and workspace in the Anthropic Console.

Why can Claude Code's cost total be higher than what I add up from my local transcript?

Claude Code can count model work that doesn't appear as a request in the main local JSONL transcript. In our October 3, 2026 test, a cost rebuilt from the transcript came in 5.65% below Claude Code's own total, mostly because of Haiku 4.5 usage during WebSearch and extra Opus requests after each turn. Read the full study →