Anthropic caps Claude power users

If you have been treating Anthropic's Claude Code like a tireless junior developer running scripts around the clock, your free-for-all is officially coming to an end.

Anthropic announced that starting August 28, it will implement new weekly rate limits across its subscription tiers. The policy applies to the $20-per-month Pro plan as well as the high-end $100 and $200 Max plans.

The clampdown targets a specific group of power users who have been running Claude Code continuously in the background 24/7, along with people who violate terms by sharing accounts or reselling access.

Double limits and API overages

Anthropic is keeping its existing five-hour usage resets in place, but it is layering two new seven-day caps on top. One sets an overall weekly limit across models, while the other specifically restricts access to the flagship Claude Opus 4.

If you subscribe to the Max tier, you will be able to buy extra headroom at standard API pricing once your weekly allotment runs out. Pro subscribers do not get that option.

Anthropic spokesperson Amie Rotherham noted that Claude Code has seen unprecedented demand since its debut, but claimed the weekly limit will affect less than 5% of subscribers based on current habits.

Behind the scenes, the limits trigger based on token consumption rather than a literal stopwatch. Anthropic estimates that Pro users can typically expect 40 to 80 hours of Sonnet 4 per week. The $100 Max plan provides roughly 140 to 280 hours of Sonnet 4 and 15 to 35 hours of Opus 4, while the $200 Max plan offers 240 to 480 hours of Sonnet 4 and 24 to 40 hours of Opus 4, delivering roughly 20 times the usage of the Pro plan.

The flat-rate agent paradox

The move follows weeks of growing pains. Anthropic had quietly tweaked limits on Claude Code recently, and its status page showed at least seven partial or major outages in the past month alone.

Anthropic is hardly alone in feeling the compute squeeze. In June, Cursor maker Anysphere overhauled the pricing on its $20 Pro plan to clamp down on power users, later apologizing for sloppy communication. Replit pushed through similar pricing changes that same month. Still, news of the new caps immediately sparked pushback across developer social media.

Anthropic says it plans to support long-running workflows with other options down the line, but the company insists these guardrails are the only way to keep the service stable right now.

Why it matters

For typical programmers who use Claude Code interactively during the workday, the 40-to-80-hour weekly ceiling on Sonnet 4 should be plenty of breathing room. The real losers are developers attempting to build autonomous background agents or multi-day pipelines on a flat $20 monthly subscription. Those workflows will now have to migrate to high-tier Max plans or raw API calls.

This shift exposes a fundamental tension in modern generative AI. Flat-rate subscriptions worked fine when humans had to manually type prompts into a chat window. But autonomous CLI tools and persistent coding agents can drain millions of tokens while you sleep, making all-you-can-eat pricing financially untenable.

We think Anthropic made the pragmatic call here, even if developers are annoyed. Experiencing seven outages a month was quickly eroding developer trust, and setting predictable weekly token ceilings is far better than unpredictable mid-day lockouts. The era of running unlimited background AI agents for twenty bucks a month is officially over, and we expect rivals to follow suit with their own weekly walls.