Upgraded Claude but usage limits did not change
You paid for a higher Claude plan and still hit the same wall. Work through the checks in order: billing timing, account mismatch, org seats, stale CLI login.
Why upgraded Claude limits do not change right away
You upgraded your Claude plan and your usage limits look the same as before. This is a support problem with a short list of causes, and every one of them is checkable from your own account. The plan change may not have completed, the client may be signed in to a different account than the one you paid for, the client may not be using your subscription at all, or the client may still be holding a session that started before the change.
Work the checks below in the order they are written. Each one names where to look and what a healthy result looks like. Do not skip to the end: the first two explain most reports.
One fact frames all of it. Every Claude surface spends from the same allowance. Anthropic's help pages state it directly:
your usage of all different Claude product surfaces (claude.ai, Claude Code, Claude Desktop) counts towards the same usage limit
So a browser tab running a long task and the claude CLI (command line interface) in your terminal draw on one pool, and a plan change applies to that one pool rather than to each app on its own. If the shape of that pool is still unclear, how Claude usage limits actually work is the background for everything here.
Check 1: did the plan change actually complete?
An upgrade is a billing change. If the payment failed and the plan did not change, restarting the client will not fix it, and the screen where you hit the limit will not always tell you that.
Open claude.ai, click your name or initials in the sidebar, open Settings, and find the billing section. Anthropic moves these labels around, so look for the section that names your current plan and shows a renewal date. A healthy result names the plan you bought, marks it active, and shows a renewal date in the future. If it still names your old plan, the change never landed, and no amount of restarting a client will fix that. If it names Pro and Pro is what you paid for, what Pro costs and where its limits sit gives you the yardstick for judging whether the allowance you are getting matches the plan you are on.
Then look at the receipt. Anthropic describes an upgrade as immediate, and its invoice guide describes the billing-cycle restart that applies to Pro-to-Max upgrades:
If you upgrade your plan in the middle of a billing cycle, you'll reset your billing cycle and receive an immediate invoice for the change rather than waiting for your next renewal.
So a Pro-to-Max upgrade starts a new billing cycle rather than finishing the one you already paid for. For that upgrade, the charge is one full cycle of the new plan, less the unused value left on your old plan, and the renewal after that is billed at the full price. Max-to-Ultra upgrades work differently: Anthropic says the charge is prorated for the remainder of your existing billing cycle. For a web subscription, an upgrade that completed produces an immediate invoice, but an account credit can cover the amount due without a new card charge. If the invoice is there but the amount is not the number you were expecting, that is usually proration, tax, or an account credit rather than a failed change, and how Claude invoices and VAT work explains where the difference comes from. If you subscribed through Google Play, the upgrade works differently: Anthropic says you are charged the full price of the new plan and the unused value of the old one is converted into extra days, so a full-price charge there is normal. For Apple App Store or Google Play subscriptions, check your app-store receipts, because Anthropic does not issue those invoices. If your paid plan is missing, check your recent payment status in Settings > Billing.
Direction matters, because billing is not symmetric here. Moving up applies at once. On the way down, Anthropic's help center spells out the cancellation side: "Your cancellation will take effect at the end of your current billing period." Anthropic's pricing page also confirms that downgrades take effect at the end of your current billing period, so check the current plan and scheduled change in your billing section, and see changing or cancelling a Claude plan for that side of the process.
Check 2: are you signed in to the account you paid for?
This is the single most common cause, and it stays invisible until you compare two strings side by side.
In the browser, open Settings and read the email address on the account. In Claude Code, run /status, which prints the account the CLI is signed in as and the authentication method it is using. Compare the two addresses character by character. They must match exactly.
They often do not match, because it is easy to end up with more than one Claude account without meaning to. Signing in with Google using a work address creates a different account from an email-link sign-up on your personal address. Both accounts work. Only one of them was charged. The other one keeps serving you its own allowance, which is the allowance you did not upgrade.
The account menu also lists workspaces, so a personal account and an organisation you belong to appear as separate entries. Which entry is active decides which entitlement applies to the request you just sent. Check that the active workspace is the one your subscription belongs to, in the browser and in the CLI, because they hold that choice separately.
Check 3: is Claude Code using your subscription at all?
A subscription only applies if the client is authenticating with it. One environment variable can change which credential it uses. Anthropic's Claude Code authentication docs explain the approval step for ANTHROPIC_API_KEY:
In interactive mode, you are prompted once to approve or decline the key, and your choice is remembered.
Claude Code's authentication docs add the step that decides it. In an interactive session, Claude Code asks once whether to use the key and remembers your answer, which the "Use custom API key" toggle in /config changes later. In non-interactive mode (claude -p), the key is always used when present. So a set variable is the lead, and /status is the proof: it shows which credential is active and marks the one that is not in use.
List what is set in your shell without printing any secret into your scrollback:
env | grep -E '^(ANTHROPIC|CLAUDE_CODE)' | cut -d= -f1If bash answers with syntax error near unexpected token and names ( instead of printing a list, the straight single quotes were probably turned into curly ones when you copied the command, which leaves the parentheses unquoted, and how to read bash's unexpected token error shows how to work back from the token bash names to the cause. cut -d= -f1 keeps the variable names and throws away the values, so you learn that ANTHROPIC_API_KEY is set without pasting the key itself into a terminal you may later screenshot for support. If it appears, find where it comes from:
grep -rl 'ANTHROPIC_API_KEY' ~/.bashrc ~/.zshrc ~/.profile ~/.bash_profile 2>/dev/null-l prints file names only, again so the key stays off your screen. Two more places set it outside your shell. Claude Code's own settings file at ~/.claude/settings.json has an env block that applies to every session, and a repository can carry .claude/settings.json that does the same for anyone who works in that directory once they trust the folder. A key you inherited from a project checkout behaves exactly like a key you set yourself. If the key you find is old or revoked, the symptom changes from the wrong limit to an outright authentication failure, and fixing Claude Code's invalid API key error explains why running /login does not clear it. If the same listing shows ANTHROPIC_AUTH_TOKEN, treat it the same way: it is a bearer token for a gateway or proxy, and Claude Code ranks it above both the API key and your subscription login. An apiKeyHelper script, which supplies an API key from a script you configure, also ranks above your subscription login, and because it is a settings key rather than an environment variable, the listing above will not show it, so rely on /status to name the credential in use. One more name in the listing deserves a look: CLAUDE_CODE_OAUTH_TOKEN, a long-lived token made with claude setup-token. It does use a Claude subscription, but it is the subscription of whichever account created it, and it normally ranks above your stored /login credentials. If you run /login while it is set, only the current session switches, and every new session reads the token again until you remove it. Anthropic profiles and federation credentials can also rank above your subscription login; /status shows a Profile row when one is active.
The same listing can also show CLAUDE_CODE_USE_BEDROCK, CLAUDE_CODE_USE_VERTEX, or CLAUDE_CODE_USE_FOUNDRY. These work differently from the credentials above, which only change what Claude Code sends to Anthropic or to a gateway. Setting any one of these three routes your requests to a cloud provider account (Amazon Bedrock, Google Cloud's Agent Platform, or Microsoft Foundry), unless a signed-in Claude apps gateway session takes precedence and uses its gateway token instead. Neither route uses your Claude subscription allowance, so its limits are not the limits you are hitting. That path is worth understanding on its own, and running Claude Code against Bedrock or Vertex explains what changes when you use it.
All of this is about Claude Code on your own machine. Anthropic's docs say Claude Code cloud sessions always use your subscription credentials, and an ANTHROPIC_API_KEY or ANTHROPIC_AUTH_TOKEN set in the cloud environment does not override them.
This failure leaves a second piece of evidence: money. If the key was in use, the work you did after upgrading was billed per token to a Console account, so check that account for usage charges before you assume nothing happened. Eligible API-key runs of claude -p can use monthly API credits in a linked Console organisation, so check credit consumption as well as charges. the difference between API billing and a subscription covers why the two feel so different in practice.
Check 4: was a usage window already running when you upgraded?
Usage limits are enforced over windows. Anthropic's Max plan page describes a five-hour session limit and says the weekly limit "resets at a fixed time each week that is assigned to your account", a reset day and time that stay the same no matter when you start using Claude or when your subscription begins. The same page says your next reset time is shown in Settings > Usage. A window that was already running when you upgraded is not, by Anthropic's account, a reason to keep hitting the old wall. As of October 27, 2026, its Max sign-up article says "Your Pro plan is upgraded to Max immediately" and "Your usage also starts fresh on the Max limits right away, so if you've reached your Pro limit, you can keep using Claude without waiting for your session to reset." Its usage-limits article states the general rule: "If you upgrade, your new plan's limits apply right away."
So upgrading in the middle of a heavy working session is exactly the case the help center describes, and the new allowance should be usable straight away rather than on the next window. What the upgrade does not change is the fixed weekly reset time assigned to your account. Verified Ultra accounts have no five-hour limit; unverified Ultra accounts have a temporary limit until Anthropic verifies them automatically.
That makes the reset time a weak test. The limit message names one, and /usage in Claude Code shows your plan usage limits, but a weekly reset that lands on the same day and time as before the upgrade is what Anthropic says should happen, so on its own it does not show that your old plan is still being applied. The signal that matters is the one the help center rules out: if you hit a limit message straight after upgrading, before you have done much new work, the request is probably not reaching your new plan. That points back to the account, authentication, and session checks: go back to Checks 2 and 3, then on to Checks 5 and 6.
More than one window can apply at the same time, so clearing the shorter one does not always clear the longer one. Reading those messages correctly is most of the skill, and what to do when you hit a Claude limit walks through them. For messages about tool calls inside a reply, see tool use inside a single turn.
Check 5: are you using an organisation seat instead of your own plan?
Team and Enterprise access is per seat, and an administrator assigns seats to people. Paying for seats does not assign them. So an organisation can hold seats while your address holds none, and you keep getting whatever your personal account provides.
The reverse trips people up more often. You upgrade your personal plan, then keep working inside the organisation workspace, so every request is authorised by the organisation's entitlement and your personal upgrade never comes into play. Switch workspaces in the account menu and try again. If the organisation workspace is the one you want, ask the administrator to confirm a seat is assigned to your exact address and that you accepted the invitation, because pending invitations already occupy seats. how Claude Team plans work for a small business covers the seat model.
Check 6: is the client session older than the upgrade?
Quit it and start it again first.
If Claude Code is waiting to continue automatically after a limit reset, it does not check for an upgrade you made separately in the browser. Sign back in after upgrading to make it check whether usage is available again.
If the limit is still wrong, sign out and sign back in. Run /logout inside Claude Code, close the terminal window, open a new one, start claude, and run /login. Do not delete ~/.claude.json by hand to force this. That file also holds your personal MCP (model context protocol) server configuration, per-project trust decisions, and interface preferences, so deleting it throws away far more than a login.
In the browser, sign out and sign in again, or open claude.ai in a private window. A private window is the faster test, because it proves whether the problem follows your account or your cached session.
While you are there, check the version with claude --version, and run /doctor, which runs a setup checkup that diagnoses installation and configuration problems. One caution on commands: as of October 2026, Anthropic documents that not every slash command appears for every user, since availability depends on your platform, plan, and environment. A missing /upgrade entry is not evidence of a broken account.
Do promotional limit increases explain it?
Temporary increases are a real category. They get announced, they run for a period, and they end. When one ends you return to your plan's normal allowance, which feels like a downgrade nobody told you about. If you run sessions in the cloud, see how cloud session credits are used and check the terms of any credit on your account.
Two things follow. Do not report the end of a promotion as a billing fault, because the account is behaving correctly. And do not size ongoing work around a temporary increase, because the work will outlive it. Check your own announcement emails and any in-app notice for the terms that applied to your account, rather than a figure someone repeated in a forum thread months later.
The order to check things in
- Billing section names the new plan as active, with a future renewal date.
- An invoice or app-store receipt exists for the upgrade.
- The email in the browser matches the email that
/statusprints in the CLI. /statusshows your subscription login as the active method. A setANTHROPIC_API_KEYalone does not mean it is in use in an interactive session.- You can keep working right after the upgrade without an immediate limit message. A weekly reset time that did not move is expected, not a fault.
- The active workspace is the one that holds the subscription.
- The client has been fully signed out and back in since the upgrade.
What to send when you contact support
Billing and entitlement problems belong with Anthropic support, not with the Claude Code issue tracker, because nobody reading a public repository can see your subscription. Attach the following, and the first reply you get will be an answer instead of a request for details:
- The email address on the account you paid for.
- The plan name your billing section shows right now.
- The date and time you changed plan, including your timezone.
- The invoice or receipt number for the upgrade charge.
- The exact wording of the limit message, copied as text rather than described.
- Which surface produced it: claude.ai, the desktop app, or the
claudeCLI. - For Claude Code, the output of
/statusandclaude --version, with any key value removed. - Whether you were in your personal workspace or an organisation workspace at the time.
- Which of the checks above you already ran, and what each one showed.
What if every check passes?
Then the plan is applied and you are simply doing more work than it covers. That is a different problem with different answers. Long sessions spend far more than short ones because each turn resends the accumulated context, so the cost of a conversation grows as it goes, which how Claude Code spends tokens breaks down. Once that is understood, the question becomes which tier fits the work you actually do, and the difference between Max and Ultra is the comparison to read before you spend more money on a limit that was never the constraint. Max 20x is now called Ultra, and existing Max 20x subscribers move to Ultra automatically with no price change.
FAQ
Does upgrading my Claude plan reset my usage limit immediately?
Yes, according to Anthropic's help center as of October 27, 2026. The upgrade applies immediately and, for web subscriptions, is invoiced straight away: a mid-cycle Pro-to-Max upgrade starts a new billing cycle and charges one full cycle of Max, less the unused value of your Pro plan. A Max-to-Ultra upgrade is instead prorated for the remainder of your existing billing cycle. Google Play instead charges the full price and converts the unused value into extra days on the new plan. On usage, the Max sign-up article says your usage starts fresh on the Max limits right away, so a Pro limit you had already reached does not make you wait for the session to reset, and the usage-limits article says a new plan's limits apply right away. Your weekly reset keeps its fixed day and time, so seeing the same weekly reset time as before is normal. If you hit a limit message straight after upgrading, take that as a sign the request is not using your new plan, and work through the account, authentication, and session checks above.
I upgraded but Claude Code still says I hit the limit. What do I check first?
Check the account, then the authentication method. Run /status in Claude Code and compare the email it prints with the email in your claude.ai settings, character by character, because a Google sign-in on a work address and an email-link sign-up on a personal address are two separate accounts. /status also shows which authentication method is active. Then run env | grep -E '^(ANTHROPIC|CLAUDE_CODE)' | cut -d= -f1. If /status confirms that ANTHROPIC_API_KEY is active, Claude Code authenticates with that key instead of your subscription, so your plan is not being used at all and the work is being billed per token.
Why does my browser show the new plan while the CLI does not?
The two clients hold their own sessions. Restart the CLI first. If nothing changes, run /logout, close the terminal, open a new one, start claude, and run /login to authenticate again with the account that holds your current plan. Avoid deleting ~/.claude.json to force it, because that file also stores MCP server configuration and per-project trust settings.
Can my organisation seat override the plan I bought myself?
Yes, and this is a normal cause rather than a fault. Requests made inside an organisation workspace are authorised by that organisation's entitlement, so a personal upgrade has no effect while that workspace is active. Switch to your personal workspace in the account menu and try the same work again. If you meant to use the organisation, ask an administrator to confirm a seat is assigned to your exact address, since buying seats and assigning them are two separate steps.