Upgraded Claude but usage limits did not change

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 usage window that was already counting when you upgraded may still be counting.

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 payment. If the payment did not go through, nothing changed, 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, with the money settled across the cycle you already paid for:

when upgrading from a lower to a higher tier, your account will be charged on a prorated basis for the remainder of your billing cycle

An upgrade that completed produces a charge and an invoice stamped with the moment you upgraded. No invoice means the change did not finish. The usual reason is a declined card, or a bank verification step that opened in another tab and was closed before it finished.

Direction matters, because billing is not symmetric here. Moving up applies at once. Moving down applies at the end of the period you have already paid for, which is the mechanism described in changing or cancelling a Claude plan. If you upgraded, changed your mind, downgraded, and then upgraded again, the state you land on can sit one period behind the state you expect.

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-and-password 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 quietly takes that decision away from you. Anthropic's Claude Code plan page is blunt about it:

If you have an ANTHROPIC_API_KEY environment variable set on your system, Claude Code will use this API key for authentication instead of your Claude subscription (Pro, Max, Team, or Enterprise plans), resulting in API usage charges rather than using your subscription's included usage.

List what is set in your shell without printing any secret into your scrollback:

env | grep -E '^(ANTHROPIC|CLAUDE_CODE)' | cut -d= -f1

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. A key you inherited from a project checkout behaves exactly like a key you set yourself.

The same check covers CLAUDE_CODE_USE_BEDROCK and CLAUDE_CODE_USE_VERTEX. Either one routes your requests to a cloud provider account, so your Claude subscription is not involved and 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.

This failure leaves a second piece of evidence: money. If a key was set, 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. 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 a window. The window starts when you start using Claude and resets at a set point after that. Changing your plan changes the ceiling that the window is measured against. It does not erase the usage the current window has already recorded, and it does not move that window's reset time.

So the reader who upgrades in the middle of a heavy working session gets the smallest possible payoff. You bought more headroom, and you spend the rest of the current window bumping into usage you had already run up. The extra room becomes visible on the next window, not on this one.

The evidence is the reset time. The limit message names one, and /usage in Claude Code reports usage for your session. If the reset time on screen was already scheduled before your upgrade, then the window you are in began before the change, and it will finish under the accounting it started with. Wait for the reset, then repeat the same kind of work and measure again.

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.

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, because an invitation that was never accepted looks identical to a seat from the outside. how Claude Team plans work for a small business covers the seat model.

Check 6: is the client session older than the upgrade?

A running client decided what you are entitled to when the session started. Quit it and start it again first, because that alone resolves a surprising number of these reports.

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 MCP (model context protocol) server configuration, per-project trust settings, and caches, 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 performs a setup checkup and flags an out-of-date install. One caution on commands: as of August 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.

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

1. Billing section names the new plan as active, with a future renewal date.
2. An invoice exists, dated when you upgraded.
3. The email in the browser matches the email that /status prints in the CLI.
4. No ANTHROPIC_API_KEY, CLAUDE_CODE_USE_BEDROCK, or CLAUDE_CODE_USE_VERTEX is set anywhere.
5. The reset time on the limit message is later than your upgrade, not earlier.
6. The active workspace is the one that holds the subscription.
7. 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 claude CLI.
• For Claude Code, the output of /status and claude --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 the two Max tiers is the comparison to read before you spend more money on a limit that was never the constraint.

FAQ

Does upgrading my Claude plan reset my usage limit immediately?

The upgrade itself applies immediately, and Anthropic bills it on a prorated basis for the rest of your billing cycle. The usage window you are currently inside is a separate thing. It keeps the usage it has already recorded and keeps its existing reset time, so an upgrade raises the ceiling without clearing what you have already spent. If the reset time on your limit message was set before you upgraded, wait for that reset before judging whether the new plan is working.

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 a password sign-up on a personal address are two separate accounts. Then run env | grep -E '^(ANTHROPIC|CLAUDE_CODE)' | cut -d= -f1. If ANTHROPIC_API_KEY appears, Claude Code is authenticating 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, and each one decided your entitlement when it started. Restart the CLI first. If nothing changes, run /logout, close the terminal, open a new one, start claude, and run /login, which issues a fresh session against 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.