Anthropic posted on X via its @ClaudeDevs account: “Starting September 14, we’re permanently raising standard weekly limits in Claude Code by 25% for Pro, Max, Team, and seat-based Enterprise plans. Until then, the current 50% increase will be in place.” Read fast, that’s an announcement about more capacity. Read slowly, it’s an announcement about less.
The math Anthropic buried in its own sentence#
Claude Code has been running a temporary 50% boost to weekly usage limits since mid-May, extended four separate times over the summer as Anthropic kept pushing the expiration date back. Call the original baseline 100 units. The summer promo took that to 150. Starting September 14, the “permanent 25% increase” takes the baseline — not the promo level — to 125.
So the actual before-and-after for anyone who’s been running at the summer’s boosted capacity: 150 units today, 125 units after September 14. That’s a 17% reduction in what’s currently available, wrapped inside a headline about a 25% increase. Both numbers are technically accurate. They’re just measuring from different starting points, and Anthropic’s own post led with the one that sounds better.
Bleeping Computer, Notebookcheck, and Android Headlines all converged on the same framing within a day of the post going up, and the reaction on X was swift enough that Anthropic followed up acknowledging the 17% figure directly rather than letting the “25% increase” framing stand uncontested.
Why this is worth a full article and not a footnote#
This blog has flagged this exact pattern from Anthropic before. Back in April, we covered Anthropic’s silent “effort” default change — a March 3 adjustment that quietly dropped Claude’s default thinking depth by 73% in one large-scale analysis, announced nowhere, discovered only through user-side measurement. The throughline across both incidents isn’t that Anthropic is uniquely bad at communicating capacity changes — every lab manages usage economics behind the scenes, and unlimited-feeling promos always have an expiration date built in somewhere. It’s that Anthropic’s own trust pitch to developers rests specifically on being the transparent, technically-honest alternative to the marketing-speak of IDE-wrapper competitors. A usage-limits post that leads with the flattering number and needs a follow-up correction to surface the actual net effect is a small instance of the exact failure mode that pitch is supposed to rule out.
To be fair to Anthropic here: this is a real, if modest, permanent increase over the pre-May baseline — 125 versus the original 100 is genuinely 25% more capacity than Claude Code shipped with before this whole promotional cycle started. Nobody’s usage is dropping below where it stood in April. The complaint isn’t that limits are being cut from a pre-promo baseline; it’s that four months of extending a “temporary” boost trained users to treat 150 as normal, and the September 14 change removes that normal without saying so plainly in the first sentence.
What to actually do about it#
If you’re a Claude Code power user who’s been running comfortably at the current boosted limit, budget for roughly a sixth less weekly capacity starting September 14. Concretely:
- Audit your current usage against the baseline, not the promo. Run
/usage(or check the dashboard at claude.com) over the next two weeks and note whether you’re regularly hitting 100–125 units versus pushing toward 150. If you’re already comfortably under 125, this change costs you nothing. - Reserve heavy agentic runs — large refactors, Dynamic Workflows-style parallel subagent fan-outs — for whichever days your weekly window resets, so you’re not caught mid-task by a limit that’s 17% tighter than it was in August.
- Consider whether Team or Enterprise seat-based plans make more sense if you’re consistently running against the ceiling; the seat-based Enterprise tier gets the same 25%/125-unit baseline, but pooled team capacity smooths out individual spikes better than an individual Pro or Max plan does.
- If cost is the binding constraint rather than the limit itself, Claude Sonnet 5’s now-permanent $2/$10 per-million-token pricing (confirmed unchanged as of mid-August) remains the cheapest path to high-volume agentic work without leaving the Claude Code default model.
None of this is a reason to reach for a different tool — every frontier lab prices usage around real GPU constraints, and a 17% adjustment after a four-month promotional run is a rounding error against what Claude Code delivers versus IDE-bound alternatives. But it’s exactly the kind of change worth reading past the headline number on, from any vendor, including this blog’s favorite one.
Sources:
- ClaudeDevs (@ClaudeDevs) on X
- Anthropic is cutting Claude Code’s current weekly limits by 17% — Bleeping Computer
- Anthropic announces a 25% increase to Claude Code limits, but there’s a 17% catch — Notebookcheck
- Anthropic’s Claude Code Usage Cap Update Actually Hides a Quiet Downgrade — Android Headlines
- Claude Code’s Weekly Limits Drop 17% on September 14 — Digital Applied
- This blog’s prior coverage of Anthropic transparency: Anthropic’s silent “effort” default change
