For the past year, the most expensive consumer version of Claude came with a clock. A subscriber paying $200 a month received a rolling five-hour window of heavy use, and when the window closed, the work stopped, whether or not the task was finished.
That clock is gone. Anthropic renamed the plan Ultra and removed the five-hour session limit for verified accounts, keeping the weekly ceiling that sits above it. The change was posted in the company’s help center and on its social accounts on Oct. 30. Existing Max 20x subscribers were moved to Ultra automatically at the same price.
The architecture of Claude’s paid tiers had two meters. A five-hour window capped how much a single session could consume; a weekly allowance capped the seven-day total. Whichever ran out first stopped the user. Removing the short meter leaves one number to manage, and Anthropic has set it at 10 times the Pro plan’s usage every week, with a separate weekly allowance for the model the company calls Fable. The $100 tier, previously Max 5x, is now simply Max. Pro remains $20.
The arithmetic of that rename is less generous than it sounds. Max 20x was sold as 20 times Pro usage per five-hour session. Ultra is sold as 10 times Pro usage per week. The unit changed from burst to budget, and the headline multiple fell by half while the price held. Anthropic says the weekly quantity itself is unchanged from what Max 20x subscribers already had. Read one way, the company standardized a limit it had been describing loosely. Read another, it repriced attention: same dollars, less peak throughput, more continuity.
Why the five-hour window existed is not mysterious. Inference capacity is provisioned against peak concurrent demand, and a small number of heavy users can consume a disproportionate share of it during business hours in California and Europe. A rolling window is a crude but effective instrument: it caps any single subscriber’s draw at a moment when everyone else wants the same GPUs. A weekly pool allows the provider to smooth consumption across days, which is cheaper to serve and easier to forecast.
For most of the people who pay $200, the constraint that mattered was never the total. It was interruption. Developers running Claude Code on a large refactor, or an agent executing a long chain of tool calls, do not consume their allowance evenly. A single intensive session can absorb a large fraction of a week’s budget in a few hours, and in the old structure it would also hit the five-hour wall, forcing a stop mid-task and a restart later. A GitHub issue opened on Oct. 15 described a related complaint: a subscriber who upgraded from Max 5x to Max 20x mid-cycle found the weekly cap had not been recalculated, so the higher tier behaved like the lower one until reset.
The dissatisfaction has already produced litigation. A developer named Karl Kahn filed a proposed class action in federal court in the Northern District of California on June 14, alleging that the $200 plan delivered roughly six to eight times the Pro plan’s usage rather than the 20 times advertised, and that a single five-hour coding session could consume about 15% of a week’s quota. The case is pending. Anthropic has described its limits as a capacity-management tool rather than a marketing promise.
Both things can be true, and the tension is structural to how AI subscriptions are sold. A software company sells features and can add seats at near-zero marginal cost. A model provider sells a share of a fixed pool of accelerators, and every pricing page is therefore an allocation policy wearing a product name. When a lab publishes a multiple, it is promising a quantity it does not fully control, because the compute behind it is bought months in advance and the demand curve is not.
Against that backdrop, the direction of the change is sensible even if the language is not. Continuity is what agentic work requires. Anthropic has spent the year pushing Claude from a chat window into an agent that runs for hours: Claude Code in the terminal, background tasks, and the merging of its chat and Cowork products into a single interface. A product that stops every five hours is a poor host for work that takes six. Selling uninterrupted time is a better match to what the company is asking customers to do with the model.
The competitive question is whether rivals follow. Every major lab meters usage in some form, and the tiers differ mostly in where the wall sits and how honestly it is described. If the market converges on weekly pools, the differentiator becomes the size of the pool and the quality of the models inside it, which is a more legible comparison than a session multiple nobody can verify.
Two things will settle whether the rename is substance or marketing. The first is whether weekly caps hold under sustained agentic load, or whether users discover that the single remaining meter bites sooner than the old pair did. The second is whether “verified” becomes a quiet gate: accounts that fail automatic verification keep a temporary session limit, and verification is decided by Anthropic from payment and account history, not by anything the customer can do at the counter.
Neither is visible on a pricing page. Both will show up in the same place the five-hour complaints did, which is the issue trackers and the forums where people who spend $200 a month on a subscription describe what happened at hour four.
