What Claude Code Pro limits actually cap, in a week Anthropic kept moving them

On July 9, the official Claude Code account posted the five words a lot of engineering teams had been refreshing their terminals for: “We’ve reset 5-hour and weekly rate limits for all users.” One day earlier, Anthropic had quietly extended Fable 5’s inclusion in paid plans by five days. Two days later, the temporary boost that lifted everyone’s weekly ceiling is scheduled to roll off. Same twenty-dollar Pro plan. Same hundred-and two-hundred-dollar Max plans. Three different answers, inside one week, to a simple question: how much can this team actually run before it gets throttled.

Three changes to Claude Code Pro usage limits in five days

Here is what moved, in order. On July 8, Anthropic pushed Fable 5’s included-in-plan window out to July 12, five more days than the earlier July 7 cutoff, with access capped at up to half of a plan’s weekly usage. On July 9, it flushed the counters, resetting the 5-hour and weekly rate limits for every Pro, Max, Team, and seat-based Enterprise user. And on July 13 at 6PM Pacific, the 50% weekly-limit boost that Anthropic switched on back in May is set to expire. Only the weekly limits revert. The 5-hour session caps that doubled in May stay where they are.

Why the weekly ceiling your team plans against keeps moving

Two things make this hard to plan around. First, Anthropic does not publish a token quota for any plan. It publishes multipliers: Pro is 1x, Max is 5x or 20x. Any guide that quotes a precise “forty-four thousand tokens per window” number invented it. Second, the limit is really two limits stacked, a 5-hour rolling window for short bursts and a weekly cap on active compute hours. The 50% boost lifted the weekly one, and it is the weekly one about to drop.

The boost was never a gift with no strings. It arrived in May as a defensive move, right as OpenAI’s Codex was winning the token-efficiency argument, and it came printed with an expiry date. The resets are the same story from a different angle: a manual lever Anthropic keeps pulling by hand while it clears a rough patch of outages. That is not a pricing page. That is a weather report.

The number a capacity plan rests on is one the vendor can move in forty-eight hours, and did, three times, this week.

Pro versus Max token limits, and the gap that matters before July 13

Here is the durable version, with the caveat that every hour figure below is a reported estimate, not a published Anthropic number. Under the boost, Pro has run at roughly 40 weekly hours of Sonnet work before the weekly ceiling bites. The Max 5x plan sits several times higher, and Max 20x higher still. When the boost expires, those weekly ceilings step back down while the price stays put. Pro’s token limit was already the tightest of the three tiers, so a reverting boost makes the Pro-to-Max gap matter more, not less, for anyone deciding who needs an upgrade this month.

The teams that feel it first are the heavy Claude Code users, the ones running long agent sessions all day, because the weekly cap is exactly the wall they hit before anyone else. For an executive, the uncomfortable part is not the price. It is that the throughput a Q3 headcount-equivalent estimate rests on can be re-set by the vendor on two days’ notice. Budgeting engineer-plus-agent output against Claude Code Pro limits means budgeting against a moving number, and it is better to know that going in than to discover it when a team’s output quietly halves in the second week of the quarter.

How to check your Claude usage limit before the reset

Three small moves before Monday. Run the /usage command inside Claude Code to see where each heavy user actually sits against the current, still-elevated ceiling. That is how to check a Claude usage limit without guessing at a token-per-day figure nobody publishes. Build a single page that lists each engineer’s tier, their real usage this week, and the date the weekly boost reverts. And name one person who owns re-checking that page after July 13, because the ceiling will have changed and the plan should move with it.

None of this is cause for alarm. Anthropic is not the only vendor turning a subscription into a meter; GitHub moved every Copilot plan onto usage-based credits on June 1. The pattern is the same across the category: the flat seat is quietly becoming a variable line. The teams that stay calm about it are the ones who know their own numbers before the vendor changes theirs. That is a very figure-out-able place to stand.