Anthropic will cut weekly Claude Code limits by 17 percent on September 14. In the official tweet, the company announced it is “permanently raising standard weekly limits in Claude Code by 25%”, only to downplay the fact that it will replace the current 50% promotional increase. Existing subscribers will lose 16.7% of today’s capacity. In its defense, that 50% boost was temporary. Only, my workflow would not survive the upcoming cut, and community reports suggest many others would not either.

In my workflow, each five-hour window represents roughly 20% of my weekly Pro limit, yet I exhaust it after only one or two hours of active work. Five windows therefore provide at most about ten working hours. Because those windows reset on a schedule, I reach the weekly limit within roughly 1.5 days. That has been my experience for the last few weeks.

What exacerbates the problem is the degraded performance of Opus 5. Anthropic disclosed its text-watermarking plans on August 14. Though it may be a mere coincidence, the failures I’ve encountered were specific: Claude repeatedly switched corrected words back or reverted sentences and their placement within paragraphs. According to the company’s own explanation, watermarking has no practical effect on output quality or content. I am not claiming its explanation to be false; rather, I think it’s more reasonable to assume its operations have taken a hit from all the recent changes it is introducing.

When Fable 5 was re-released, I subscribed to Max 5x for a month to try it. In my experience, Fable 5 at high effort produced output comparable to Opus 5 only at extra high. That is why I now use Opus 5 at extra high. It is the best compromise I have found between conserving my limit and reaching that quality.

The cut will also apply to Max 5x, the cheapest individual subscription with included Fable usage. It may matter even more there than on Pro because Fable usage is capped at half of the weekly limit and Anthropic states that Fable consumes the allowance faster. My annual Pro plan is ending either way. I am considering an “upgrade” to Max 5x for one month because Anthropic would convert my remaining Pro balance into subscription credit. If the service does not improve, I would leave any remaining credit for a future subscription when I need Claude as a secondary agent or until Anthropic changes its pricing model again.

Community reports of faster drain and degraded performance across Pro and Max are numerous but still anecdotal; I know of no reliable benchmark showing either problem to be widespread. That uncertainty does not change the effect of the September cut. It guarantees a lower ceiling on a service whose limits are already opaque and, in my experience, increasingly difficult to work within. OpenAI has distributed complimentary resets that users can bank and redeem later. Anthropic has no such bank: if a complimentary reset lands on a customer’s regular weekly reset, as mine did, the two coincide and one benefit is effectively lost.

Perhaps Anthropic has something in store in 2 weeks to significantly enhance usability of its models. Or perhaps it will provide more detailed stats and/or tooling on each user’s actual token usage. Otherwise, if push comes to shove, I don’t see why anyone would want to trust a subscription model that returns 17% less.