On May 13, Anthropic quietly gave Claude Code users a 50% usage boost. Temporary, they said. A promo. The kind of thing that ends in a few weeks. Except it didn't. They extended it. Then extended it again. By the time it ran through four extensions over the summer, the boosted rate had been active for more than three months. At that point, the boosted rate isn't the promo. It's the baseline developers have tuned their workflows against.

Then last week, ClaudeDevs posted on X: "Starting September 14, we're permanently raising standard weekly limits in Claude Code by 25%." Positive framing. Real claim. And then the developer community did the math.

The 25% increase compares the new permanent level to the original standard baseline. New = 125% of standard. But developers today are running at 150% of standard — the boosted rate. The permanent increase lands at 125. The current meter reads 150. That's 125 ÷ 150 = 0.833. A 16.7% reduction from today's headroom. Both the 25% raise and the 17% cut are mathematically correct. They're just measuring from different places.

Anthropic deleted the original post and republished with corrected framing. That's the right call.

17%

→ Source: ClaudeDevs / Anthropic

Source spread

• ClaudeDevs on X [hype] — Led with the 25% raise, no mention of the current boosted-rate delta. The official announcement.
• Notebookcheck [skeptic] — "25% increase, but there's a 17% catch." Gets the math right.
• BleepingComputer [skeptic] — "Anthropic is cutting Claude Code's current weekly limits by 17%." Accurate from the developer's lived-experience baseline.
• explainx.ai usage limits timeline [builder] — Full dated history of every Claude Code limit change since May 13. Useful for understanding the full extension count.

Pros & cons

What's true and useful:

• The permanent baseline goes up 25% on September 14. Real, lasting change. You end up better off than before May 13.
• The corrected post is available. Anthropic acknowledging the framing gap is better than doubling down on misleading math.
• Usage doesn't translate to a fixed prompt count. Weekly limits are token-weighted and vary by conversation length, model, tool use, and reasoning depth. The 17% figure is an approximation — not a hard ceiling reduction.

What deserves a side-eye:

• Running a "temporary" promo through four extensions across nearly five months, then framing its end as a net positive, is a communications failure of Anthropic's own making.
• The original announcement required external press to correct it before Anthropic revised. That's not a great look for a company where "builder trust" is a core value proposition.
• Developers who calibrated workflows against the boosted rate have real recalibration work ahead, and that context was entirely missing from the initial announcement.

Samwise's take

What builders need to know

• September 14 is the hard date. After that: permanent 25% over the standard baseline for Pro, Max, Team, and seat-based Enterprise plans.
• Today's rate is 50% over standard. September 14 is 25% over standard. From your current meter, that's a 17% reduction.
• Token consumption is usage-pattern-specific. High-tool-use, long-context, or heavy reasoning workflows will feel the 17% more than lighter sessions.
• The new permanent level is still better than pre-May 13. Five months of preview headroom that you don't get to keep. September 14 lands between "pre-promo" and "peak boost."
• Budget conservatively until October. First billing cycle after September 14 is your first real data point on the new limits. Don't assume the old numbers hold.

Further reading

• ClaudeDevs on X — the original announcement
• Notebookcheck — "25% increase, but there's a 17% catch"
• BleepingComputer — "Anthropic is cutting Claude Code's current weekly limits by 17%"
• Claude usage limits 2026: full dated timeline