Since May 2026, Anthropic has run a temporary +50% boost on Claude Code’s weekly usage limits across Pro, Max, Team and seat-based Enterprise plans — extending it several times, most recently through 13 September 2026. On 14 September, that boost disappears and a new “permanent” limit takes its place: 25% higher than the original, pre-boost baseline.
That framing is technically accurate and practically misleading. Compared to the baseline from before May, 25% is a genuine increase. Compared to what teams have actually been working with under the +50% boost for the past four months, the new number is a 17% cut in weekly capacity — Anthropic confirmed the reduction directly in a follow-up post. Using a baseline of 100 units: today, with the boost, a team gets 150 units a week; from 14 September, they get 125.
Why this matters if you’re not the one prompting
If your team, or an agency you’ve commissioned, is running Claude Code — or planning to — as part of an AI-assisted build, this is a capacity-planning problem, not a pricing footnote. Weekly limits govern how much autonomous agent work a plan can absorb before hitting a wall mid-sprint: refactors, multi-file changes, long-running background tasks. A 17% drop in available capacity, landing with one day’s gap between the old boost ending and the new baseline starting, is exactly the kind of change that catches a project mid-flow if nobody on the team is tracking it.
It’s also a sign that “included in your plan” numbers for AI coding tools are moving targets right now, not fixed costs. GitHub Copilot cut its included AI credits by up to 44% on Business and Enterprise plans on 1 September 2026. Claude Code’s own weekly limits have been extended, boosted, and now quietly reduced, all within the same year. Every major vendor is still working out what agentic coding at scale actually costs them to run, and adjusting what they hand to users as they go — usually with the increase in the headline and the cut in the fine print.
So what
If you’re commissioning software built with AI-assisted development, ask whoever is building it which plan tier they’re running their AI coding tools on, and whether current usage fits inside the new weekly limits — not the boosted ones from last month. A team that quietly runs over its allowance mid-project either eats the overage cost or slows down without saying so, and neither shows up in a quote. At BuildApps, tool cost and capacity are part of how we scope any AI-assisted build from the start — see our AI-assisted development work, or get in touch for a realistic estimate of what a project actually costs to build with these tools today.