Claude Code's standard weekly limits are scheduled to rise 25% on September 14, 2026, but that is about 17% less than the temporary allowance available today. Both statements are true: the current promotion gives you 150% of the old standard, while the announced new standard is 125%.
The change concerns weekly Claude Code capacity. It does not increase the five-hour limit, give Claude chat the same promotional multiplier, or turn a monthly subscription into a dollar balance. If you are already blocked, the useful question is which limit you reached and when that particular window resets.
This guide was checked on September 8, 2026. September 14 is an announced change, not an account rollout we have measured. The earlier August 31 deadline has been superseded by the extension in Anthropic's official announcement.
What changes on September 14?
The following comparison sets the old standard weekly Code allowance to 100. These are index values, not tokens, messages, coding hours, or dollars.
| Period | Weekly Code allowance, relative to the old standard | What to use when planning |
|---|---|---|
| Before the promotion began on May 13, 2026 | 100 | The baseline behind the announced percentage increases |
| May 13 through September 13, 2026, at 11:59 p.m. PT | 150 | The temporary 50% increase, still current on September 8 |
| From September 14, 2026, as announced | 125 | A permanent 25% increase over the old standard, with less capacity than the promotion |
The Help Center promotion terms give the current end date, even though the page title still says “May–August.” The official announcement supplies the new 25% standard. Anthropic's follow-up also explicitly describes the reduction from the promotional level as approximately 17%.
The promotion covers Pro, Max, Team, and legacy seat-based Enterprise plans. It applies to Claude Code in the CLI, IDE integrations, desktop, and web; it excludes Free and consumption-based Enterprise. The increase is automatic. Chat and Cowork limits and the five-hour limit are unchanged by this promotion.
The cutoff uses Pacific Time. On September 13, Pacific Time is PDT, so September 13 at 11:59 p.m. PT is September 14 at 06:59 UTC. That is a promotion boundary, not a promise that every account's weekly meter resets at that instant. The cited announcement and terms do not explain how an already-running weekly window will be handled at the transition.
How does a 25% increase become a 17% reduction?
The comparison changes its denominator. Using the table's index:
- Compared with the old standard:
(125 − 100) ÷ 100 = 25%more. - Compared with the current promotion:
(125 − 150) ÷ 150 ≈ −16.67%.
The loss is 25 index points, but it is not 25% of the capacity you have today. It is one-sixth of today's promotional capacity.
For workload planning, the inverse comparison is useful. A fixed amount of Code work takes a larger share of the smaller allowance: 150 ÷ 125 = 1.2. For example, work that consumes 80% of a full promotional week would consume about 96% of the announced new weekly allowance: 80% × 1.2 = 96%.
That example assumes the same plan, Code-only work, model mix, and consumption rules. It is a budget estimate, not a prediction of how Anthropic will redraw an existing progress bar. Under those assumptions, a workload above roughly 83.3% of a full promotional week would no longer fit within one new-standard week. If your week mixes chat and Code, or your next task uses a different model or much longer context, multiplying the observed percentage by 1.2 is not enough to forecast it reliably.
There is no published universal absolute quota in this announcement that lets you turn these percentages into a guaranteed number of prompts or hours.
Which limit stops you first: five-hour or weekly?
Whichever applicable allowance runs out first can stop the next request. The windows overlap; one is not a reserve that automatically takes over when the other empties.
Consider two illustrative account readings:
- Five-hour usage at 100%, weekly usage at 40%: weekly room remains, but the session limit is the blocker. Read the session reset time.
- Five-hour usage at 20%, weekly usage at 100%: the shorter window has room, but the weekly limit is the blocker. A five-hour reset alone will not restore weekly capacity.
Open Settings > Usage and read both the progress and reset time for each applicable row. Anthropic's usage best practices identifies this as the place to monitor session and weekly consumption. Use the actual labels shown on your account rather than assuming every plan has identical model rows.
In Claude Code, /usage also displays account plan bars. If it says Showing last-known usage, those bars are cached; press r to retry before relying on them. The local breakdown selected with d or w covers activity on that machine, not all devices or claude.ai. It cannot reconcile the whole account by itself. See the CLI usage documentation.
| Blocking message or condition | What it tells you | What to check next |
|---|---|---|
| Session limit | The total session window is exhausted | Its displayed reset time |
| Weekly limit | The total weekly window is exhausted | Its displayed reset time, even if the session bar has room |
| Opus or Sonnet limit | A model-family allowance is exhausted | Whether another family has capacity and the total limits still have room |
| Context or auto-compaction warning | The conversation is approaching its length limit | What context the task needs to retain |
| API rate-limit error | API access is hitting a separate constraint | The authenticated API account and provider's limits |
Switching models does not clear a total session or weekly cap. It can help with a model-specific cap if an eligible alternative still has room. That distinction is documented in Claude Code's limit diagnosis. For an API 429, continue with the separate Claude API rate-limit guide.
Do Claude chat and Claude Code share the same allowance?
Yes, Pro and Max usage across Claude and Claude Code counts against shared plan limits. Using the same account in an IDE does not create a separate allowance. Anthropic explains this in its Pro and Max plan guide.
However, shared limits do not mean every product receives the same promotional boost. The temporary increase is specifically for Claude Code; the promotion terms leave chat and Cowork limits unchanged. You therefore cannot treat a mixed week of chat and coding as one pool that was universally multiplied by 1.5 and will universally become 1.25. The public information does not expose the debit weights needed for that calculation.
This also explains why checking only your laptop's coding history can be misleading. To understand remaining capacity, use the account-level bars; to investigate what your local Code workflow consumed, use the local breakdown.
Before applying subscription arithmetic, confirm how the client is authenticated. Anthropic warns that setting ANTHROPIC_API_KEY can make Claude Code use API authentication instead of the usage included in a Pro or Max subscription. That work follows API billing and limits, even if you also pay for a subscription.
Does /limit-reset reset the weekly limit?
Do not count on it to restore weekly capacity. A first-person report of /limit-reset describes a session reset while the weekly limit still applies. Other users report that the option is unavailable.
Those are user reports, not a confirmed universal entitlement. The consulted official documentation does not establish eligibility, availability for every account, or a guaranteed reset frequency. If your account offers a reset, read what it actually resets and then check the remaining weekly allowance. A restored session window cannot, on that evidence, be treated as a fresh week.
There is a separate documented feature for waiting until a usage limit naturally resets and then continuing, available for supported interactive subscription sessions from Claude Code v2.1.234. Waiting for the reset does not bring it forward.
Do not assume weekly limits reset on Monday, on your subscription renewal date, or at the September 14 policy boundary. The reset time displayed for your account is the actionable one. Before pausing, save the current goal, changed files, last verification result, and next step so that resuming does not require reconstructing the task.
Why does it look as if $200 disappeared in a few days?
First identify what the $200 represents. These three observations mean different things:
| Where the number comes from | What it means | What would establish that money was spent |
|---|---|---|
| A $200 monthly subscription payment | Payment for a plan with included usage and limits | The subscription invoice; the amount is not a prepaid API balance |
| The Session cost estimate in Claude Code | A local estimate of token cost | It does not establish an extra subscription charge |
| Usage credits or API billing | Separately metered paid usage | The relevant account's usage and billing records |
If you paid $200 for a month and reached the weekly cap after several days, you have exhausted a usage window; that does not mean Claude debited $200 of API credits. Likewise, a $200 local cost estimate does not prove that an extra $200 was charged. The cost documentation distinguishes that estimate from billing.
If actual usage credits fell by $200, inspect the account's recorded spend and spending limit. If the client used API authentication, inspect the API billing record. Keep the subscription payment, included-usage percentage, and separately billed spend as three distinct numbers when comparing days or asking support for help.
The pace of allowance use also depends on the work. Anthropic's usage and length limits explanation identifies conversation complexity, model choice, effort, and features as factors. “A few days” does not describe how much work ran during those days, so another subscriber's elapsed time is not a reliable allowance benchmark.
How should you plan the first week after September 14?
Use a recent, representative Code-only week as a starting point. If it used 80% of the promotional allowance, budget roughly 96% under the announced new standard, then leave room for changes in the actual task. Do not assume a week with unusually short edits predicts the next week's large debugging job.
Before starting a long session, check both the five-hour and weekly bars, confirm that the readings are fresh, and note their separate reset times. Keep conversations focused on the current task and carry forward a concise handoff when moving to unrelated work. These habits help control consumption, but they do not reset a quota or guarantee a fixed number of hours.
After the policy boundary, verify the allowance and reset information actually displayed on your account. The announced comparison is clear—125 instead of the promotional 150—while the handling of a week already in progress remains unspecified. Plan the workload around the published ratios and use the live account counters to decide whether the next request can run.
