Claude Code Limits Are Changing Sept 14: The Real Math

The Claude Code weekly limit change officially takes effect on September 14, 2026, and from that date Pro, Max, and Team users face a different quota picture. Announced by Anthropic on August 29, 2026, this transition was framed as a 25% permanent-baseline increase — but because the temporary 50% bonus in place since May 13 ends the same day, it actually amounts to roughly a 17% reduction relative to what you can use today. This article walks through where the numbers come from, who is affected, and how to manage your quota before and after September 14, based on primary sources.

💡 Pro Tip: Before September 14, run the /usage command and note today's consumption curve — that's the only reference point that lets you compare what actually changes once the permanent regime kicks in.

Table of Contents

• What Exactly Changes on September 14
• Timeline
• Two Baselines, Two Correct Percentages — 25% Increase or 17% Cut
• Who's Affected, Who's Out of Scope
• Don't Confuse the 5-Hour Session Limit With the Weekly Limit
• Fable Models Work Differently: The 50% Threshold and Usage Credits
• Plan × Fable Behavior
• Measuring Your Limit: The /usage Command and Reading the Usage Panel
• 6 Practical Settings That Stretch Your Quota
• 1. The right model for the right job
• 2. Adjust thinking depth per task with `/effort`
• 3. Delegate to subagents
• 4. A cache-friendly CLAUDE.md structure
• 5. Manage your `/clear` cadence
• 6. Keep verbose work in a subagent, not the main thread
• Pre-September 14 Checklist
• FAQ
• How much are Claude Code's weekly limits changing on September 14, 2026?
• Why does a permanent 25% increase amount to a 17% decrease relative to today?
• Which plans are affected by this change?
• What happens when my Claude Code weekly limit fills up?
• Is the 5-hour limit the same thing as the weekly limit?
• Are Fable models excluded from my weekly limit?
• Conclusion
• Sources

What Exactly Changes on September 14

Anthropic has been granting Claude Code users a temporary quota bonus since May 13, 2026: your weekly usage limit was 50% higher than the pre-campaign baseline. That temporary increase ends at 23:59 Pacific Time on September 13, 2026, and in the support documentation's own words, "after September 13, weekly usage limits in Claude Code return to standard levels. There's no change to your plan or billing."

But "standard level" here does not mean the old (pre-campaign) baseline. As Anthropic explained in its August 29 announcement, the new permanent baseline that takes effect on September 14 is 25% higher than the original baseline. So there are two different reference points: (1) the original, pre-campaign baseline, and (2) the temporary, 50%-inflated baseline you're using right now.

The second post made the same day from Anthropic's @ClaudeDevs account summarizes this as: 'Compared to today, this works out to a 17% reduction in weekly limits on Claude Code.' (source) implicator.ai's analysis gives a worked example with a baseline of 100 units: if the pre-campaign baseline is 100, the temporary campaign raised it to 150 (today's level); on September 14, the new permanent baseline drops to 125. Caveat: implicator.ai explicitly states these 100→150→125 figures are not Anthropic's actual metering unit, but a representative illustration. Anthropic doesn't publish its weekly limit as a concrete "unit" count; the limit is a budget that varies with model mix, conversation length, and tools used.

The bottom line: keep two facts straight — (1) 125 is 25% more than 100; (2) 125 is roughly 16.7% (rounded to 17%) less than 150. Both are mathematically correct — it just depends which baseline you're measuring against.

Timeline

Two Baselines, Two Correct Percentages — 25% Increase or 17% Cut

The reason this change generated so much discussion wasn't the math — it was the framing. The first post in the same announcement thread from Anthropic's @ClaudeDevs account led with "the permanent baseline is going up 25%" — technically correct, but incomplete on its own, because what users will actually feel is a decrease relative to the current (campaign-boosted) level. In the second post of the same thread, Anthropic explicitly stated the 17% reduction relative to today as well: "Compared to today, this works out to a 17% reduction in weekly limits on Claude Code."

The practical implication: if you're a heavy user before September 14 (consuming most of your weekly limit), you're more likely to hit the wall earlier with the same workload after September 14. If you've been a light user since before the campaign, the permanent baseline is already 25% wider than it used to be, so you may not notice a difference.

You can estimate your own situation with simple arithmetic:

1// Representative example — not Anthropic's actual metering unit,
2// just a simplified calculation to illustrate the relationship.
3const baselineBefore = 100; // original, pre-campaign baseline
4const todaysLevel = baselineBefore * 1.5; // 150 — temporary 50% bonus (through Sept 13)
5const permanentLevel = baselineBefore * 1.25; // 125 — permanent baseline from Sept 14
6 
7const permanentIncrease = (permanentLevel / baselineBefore - 1) * 100; // 25.0%
8const cutVsToday = (1 - permanentLevel / todaysLevel) * 100; // ~16.7%
9 
10console.log({ permanentIncrease, cutVsToday });
11// { permanentIncrease: 25, cutVsToday: 16.666666666666664 }

Who's Affected, Who's Out of Scope

According to the support documentation, this change covers the following plans:

• Pro plan users
• Max plan users (including 5x and 20x)
• Seats on the Team plan
• Legacy seats on seat-based Enterprise plans

Who's excluded is just as clear: Free plan users aren't subject to this weekly limit mechanism at all, and consumption-based Enterprise contracts are unaffected by this change either — they already run on a pay-as-you-go model with no weekly "included" limit. Source: support.claude.com/en/articles/15910845.

If you're an admin managing multiple seats on Team or Enterprise, explain this distinction to your team ahead of time: within the same organization, some seats can be seat-based (in scope) while others are consumption-based (out of scope). A developer who doesn't know their own seat's category could run into an unexpected limit mid-week — so verifying seat type in advance via the /usage output or the billing page is the cheapest way to avoid a surprise.

Don't Confuse the 5-Hour Session Limit With the Weekly Limit

Two separate counters run in parallel in Claude Code, and mixing them up is the most common way people mismanage their quota:

• 5-hour session limit: Caps how much you can use within a given time window; once you hit it, the counter resets within five hours. Source: support.claude.com/en/articles/12429409 — "Your plan's included usage limit resets every five hours once you reach it. Usage credits don't affect this reset schedule."
• Weekly limit: The subject of this article — the limit whose regime changes on September 14. It caps total usage over a longer window (one week).

The support documentation explicitly states these two mechanisms are independent: "5-hour usage limits are not affected by this promotion." So the September 14 change only affects the weekly ceiling; the five-hour rhythm operates exactly as before. In practice, you might hit your 5-hour limit on a heavy day while your weekly limit is still far from full, or the reverse. They're two indicators you need to track separately.

Fable Models Work Differently: The 50% Threshold and Usage Credits

Since Fable models (Fable 5, Fable 5.1) became selectable in Claude Code, there's an extra layer to the weekly limit discussion: how Fable models count against your plan works differently than Sonnet or Opus.

The support documentation (support.claude.com/articles/15424964) states the rule clearly: on Max plans, on premium seats within Team plans, and on premium seats within seat-based Enterprise plans, up to 50% of your weekly usage limit can be used on Fable models; cross that threshold and you either continue on usage credits or switch models. One nuance: this 50% is not extra capacity on top of your plan limit — it's the maximum share of that same weekly ceiling that can go to Fable. In the documentation's own words: "You can use up to 50% of your weekly limit on Fable models, but usage of other models counts against the same usage limits, and you can never use more than your weekly limit."

For Pro plan users and standard seats on Team/Enterprise, the situation is simpler and less advantageous: Fable models run on pay-as-you-go usage credits from the very first token, because these plans' included usage limit doesn't apply to them at all.

Access to Fable models is also gated by a version threshold: at least 2.1.170 for Fable 5, at least 2.1.255 for Fable 5.1 (source: support.claude.com/articles/15424964). On an older CLI version, you may not see the Fable options at all — updating should be your first checkpoint.

There's also a historical distinction worth keeping straight: Anthropic previously ran a separate promotion through July 2026, letting you use up to 50% of your weekly subscription on Claude Fable 5 at no extra charge. That ended on July 19, 2026, and the support page states Claude Fable 5.1 was never part of it. So today's "50% threshold" is a different mechanism: not a free bonus, but an upper cap within the same weekly limit.

Plan × Fable Behavior

Measuring Your Limit: The /usage Command and Reading the Usage Panel

Rather than guessing, the most reliable way to see your quota directly is a command you run in the terminal: /usage. The support documentation summarizes it this way: "You can run /usage in the CLI to see your updated limits." You can also reach the same data through the web interface at claude.ai/settings/usage.

1# Run inside the Claude Code CLI
2/usage

The usage credits mechanism applies both to Claude conversations and to Claude Code terminal usage, and the support documentation clarifies: "Usage credits apply to both Claude conversations and Claude Code terminal usage. They count against limits for combined usage across both interfaces." In other words, browser chats with claude.ai and your terminal Claude Code sessions share the same weekly pool — you can burn through one thinking you're "barely using it," while hitting the limit on the other.

Rather than guessing when your weekly limit resets, the timestamp in the /usage output is a more reliable reference than the documentation's general seven-day window description.

6 Practical Settings That Stretch Your Quota

Knowing your weekly limit will fill up faster after September 14 than it does today, here are the real control points you have to reduce your consumption:

1. The right model for the right job

Instead of handing every task to the strongest (and most quota-hungry) model, delegate exploration, file scanning, and mechanical work to lighter models. Reserve the strong model for complex, clearly-specified code writing — not grep searches, file listings, or simple translation.

2. Adjust thinking depth per task with `/effort`

Instead of allocating maximum thinking budget to every request, lowering effort to match the task size directly reduces token consumption — a simple file edit doesn't need high/max effort.

1# Example: lower effort before a mechanical task
2/effort low

3. Delegate to subagents

Rather than bloating the main conversation's context with large research outputs, log dumps, or test results, having a subagent do that work and returning only the summary disciplines both context and quota consumption.

4. A cache-friendly CLAUDE.md structure

Organizing your project instructions (CLAUDE.md) so they don't change often makes it easier for prompt caching to kick in — cached portions get reused more cheaply instead of being reprocessed from scratch every session. Keep permanent rules at the top of the file, not frequently-changing "current status" notes.

Keeping the project-specific, stable parts at the top of .claude/settings.json is advantageous for the same reason:

1{
2  "permissions": {
3    "allow": ["Bash(git status)", "Bash(npm test)"]
4  }
5}

5. Manage your `/clear` cadence

Resetting with /clear when switching to an unrelated task keeps you from carrying over bloated context. When continuing the same task, narrowing context with /compact <focus> is more efficient than starting from scratch.

6. Keep verbose work in a subagent, not the main thread

Rather than dumping bulky output — build logs, long test logs, documentation scans — into the main conversation, having a subagent do it and returning only a compact summary lowers both context and, indirectly, quota consumption.

Pre-September 14 Checklist

• Run /usage and note today's (campaign-boosted) consumption level.
• Verify whether your plan is in scope for this change (Pro/Max/Team/seat-based Enterprise = in scope; Free/consumption-based Enterprise = out of scope).
• If you use a Fable model, check your Claude Code version (Fable 5 → 2.1.170+, Fable 5.1 → 2.1.255+); update if you're not current.
• Track the 5-hour limit and the weekly limit separately — they reset independently.
• During heavy-usage weeks, review the usage credits purchase process in advance (available on Pro and Max plans).
• If you work with a team (Team/Enterprise), clarify which seats are standard and which are premium — the Fable 50% rule applies only to premium seats.

FAQ

How much are Claude Code's weekly limits changing on September 14, 2026?

Starting September 14, 2026, Anthropic is permanently raising Claude Code's weekly usage limit by 25% relative to the original (pre-campaign) baseline. But the temporary 50% increase that's been in effect since May 13, 2026 ends the same day; so for current users, the net result is roughly a 17% decrease relative to today's level (Source: support.claude.com/en/articles/15910845; implicator.ai, Anthropic's August 29, 2026 announcement).

Why does a permanent 25% increase amount to a 17% decrease relative to today?

If the baseline usage allowance is taken as 100 units: the temporary May–September campaign raised it to 150 (today's level). On September 14, the new permanent baseline becomes 125 units — 25% more than the pre-campaign 100, but roughly 17% less than today's 150. Anthropic stated this explicitly in the second post of the same announcement thread (Source: implicator.ai; ClaudeDevs announcement; note: the 100/150/125 figures are, per implicator.ai's own statement, a representative calculation example, not the actual metering unit).

Which plans are affected by this change?

The change covers Pro, Max, Team, and seat-based Enterprise plans; Free plan and consumption-based Enterprise contracts fall outside this weekly limit mechanism. Fable models (Fable 5/5.1), meanwhile, count within the plan's weekly limit up to 50% on Max and premium seats; usage beyond that threshold requires either switching to another model or continuing on usage credits (Source: support.claude.com/en/articles/15910845; support.claude.com/articles/15424964).

What happens when my Claude Code weekly limit fills up?

When the weekly limit fills up, Claude Code stops usage for that week; this is independent of the separately-running 5-hour session limit — you could exhaust one while still having capacity left in the other. You can see when each limit fills up via the /usage command or claude.ai/settings/usage. Pro and Max plan users can continue past their included usage limit by purchasing usage credits at standard API rates (Source: support.claude.com/en/articles/12429409).

Is the 5-hour limit the same thing as the weekly limit?

No, they are completely separate mechanisms. The 5-hour limit is a short-term counter that resets every five hours once you reach your included usage limit; the weekly limit is a separate ceiling that caps total usage over a longer window. The support documentation explicitly states that the September 14 change only affects the weekly limit, and that 5-hour limits are not affected by this promotion (Source: support.claude.com/en/articles/15910845).

Are Fable models excluded from my weekly limit?

No, quite the opposite — on Max and premium seats, Fable models count within the same weekly limit, but can be used for up to 50% of it; this isn't extra bonus capacity, it's an upper cap within the existing ceiling. On the Pro plan and on standard seats, Fable usage runs on usage credits from the start, since it isn't part of the plan limit at all (Source: support.claude.com/articles/15424964).

Conclusion

What happens on September 14, 2026 is neither a pure "cut" nor a pure "improvement" — it's both, depending on your vantage point. The permanent baseline really is 25% wider than the pre-campaign level, but for current users who've been living with the temporary 50% bonus, it amounts to roughly a 17% contraction relative to today. Once you've internalized that distinction, the one thing left to do is plan your quota by task type: reserve complex work for the strong model and high effort, and delegate everything else to lighter models and low effort.

If you want a deeper strategy on model selection and prompt caching, read Claude Prompt Caching: Cutting Costs 10x; to learn how to parallelize work across subagents, read Claude Code Multi-Agent Teams: Working in Parallel; for planning discipline, read Claude Code Plan Mode: Architectural Planning. If you want to automate repetitive tasks with hooks, Claude Code Hooks Automation Guide can help; and if you're curious how the rest of the Fable family compares to other models, check out Claude Haiku 4.5: Speed and Low-Cost Guide.

Sources

• Claude Code May–August 2026 Weekly Limits Promotion — Anthropic's official support article; the promotion's scope, end date, and the "return to standard levels" language.
• Claude Fable Models on Your Plan — the 50% threshold for Fable models by plan, usage credits rules, and Claude Code version thresholds.
• Usage Credits — the usage credits mechanism, the 5-hour reset behavior, and how it applies to Claude Code.
• Anthropic Cutting Claude Code Weekly Limits by 17% Starting September 14 — a detailed analysis of the August 29 announcement, direct quotes from Anthropic's X statement, and the 100→150→125 worked example.
• Anthropic's "25% increase" to Claude Code limits is really a 17% cut — Enterprise DNA's independent confirmation of the framing debate.
• ClaudeDevs (Anthropic's official account) — September 14 announcement — source of the statement: "Starting September 14, we're permanently raising standard weekly limits in Claude Code by 25%... Until then, the current 50% increase will be in place."