Claude Code's Weekly Cap Drops 17% Today as Anthropic's Temporary Boost Expires

Anthropic's +50% temporary boost on Claude Code's weekly limit expired at 11:59 PM PT on September 13, and the permanent replacement that took effect September 14 leaves subscribers with 16.7% less headroom than they had on Friday. Anthropic's own follow-up post on August 29 put it more plainly: "Compared to today, this works out to a 17% reduction in weekly limits on Claude Code."

The framing gap between what Anthropic emphasized and what developers experience is the whole story. Measured against the pre-promotion baseline, the new ceiling is 25% higher and reads as a generous permanent lift. Measured against last week, it's a cut. Digital Applied laid out the arithmetic on an index scale: the old baseline sits at 100, the May 13 promotion pushed it to 150, and the new steady state lands at 125. That's 125 ÷ 150 = 0.833, and the round trip back up to 150 would require a 20% increase, not the 17% Anthropic quoted.

None of the absolute token figures are public. Digital Applied notes that Anthropic publishes no raw numbers for either meter, so every ratio in circulation is proportional by necessity. BleepingComputer relayed the vendor's own caveat that weekly allowance varies with conversation length, model choice, tool usage, and effort level.

The architecture matters. Claude Code enforces two meters: a five-hour rolling session limit and the weekly cap. As Implicator points out, a subscriber can hit the weekly ceiling even when a fresh five-hour window is available. Plan prices, five-hour limits, and Claude web, desktop, and Cowork limits are untouched, per AI Catchup's read of the rewritten Help Center article. To check current usage, run /usage in the Claude Code CLI; a GitHub issue filed August 20 documented the underlying JSON keys rate_limits.seven_day.used_percentage and resets_at.

The competitive subtext arrived on schedule. OpenAI core products lead Thibault Sottiaux said on August 31 that Codex Pro 20x delivers "quite precisely" 20 times the weekly usage of Plus with no five-hour sub-limit, a vendor claim Implicator flagged as unverified by independent measurement. The developer mood is legible in a single Reddit post Implicator surfaced from user ZombieBallz: "I never really feel like I'm watching my Pro 20x usage on Codex, but with Claude I am checking it like every 30 minutes and rationing my week out."

The rollout itself telegraphed the sensitivity. AndroidHeadlines reported that Anthropic deleted the original announcement thread and reposted a clarification after developers appended a Community Note. Cap changes are pricing changes, and in the coding-assistant market pricing changes now propagate at the speed of a screenshot.

Sources

• https://www.bleepingcomputer.com/news/artificial-intelligence/anthropic-is-cutting-claude-codes-current-weekly-limits-by-17-percent/
• https://www.digitalapplied.com/blog/claude-code-weekly-limit-reduction-september-14
• https://www.androidheadlines.com/2026/08/anthropic-claude-code-weekly-limits-update.html
• https://www.implicator.ai/anthropic-claude-code-weekly-limits-september-14/
• https://aicatchup.com/news/claude-code-weekly-limits-permanent-25-percent-september-2026