Anthropic Admits Claude Code Quotas Are Draining 'Way Faster Than Expected' — Cache Bugs, Peak-Hour Cuts, and a Promo Expiry Collide (HN: 259pts, 158 comments)
While the source code leak dominated developer Twitter on March 31st, a different crisis was unfolding on developer Discord: Claude Code users were hitting their usage limits in minutes rather than hours. Anthropic has now officially acknowledged the problem, with a Reddit statement confirming the issue is under investigation at "top priority." The Register's reporting identifies three compounding causes that collided in the same week. First, a March 26th quota reduction during peak hours hit harder than the stated 7% of users. Second, a promotion that had been doubling usage limits outside peak windows quietly expired on March 28th — creating an effective quota halving for users who didn't notice. Third, and most technically significant: two independent prompt cache bugs are causing cache to fail silently, inflating effective token costs by 10 to 20 times.
The real-world impact is stark. A Max 5 subscriber ($100/month) reported exhausting their quota in one hour instead of the expected eight. A Pro plan user ($200/year) is hitting limits every Monday and getting usable access on only 12 of 30 days in a month. For teams doing serious agentic work — the exact users Claude Code is designed for — this kind of quota volatility is a fundamental reliability problem. A developer who reverse-engineered the Claude Code binary before the npm leak (an independent discovery, notably) identified the cache bugs; downgrading to v2.1.34 provided noticeable relief for multiple users. The fix, however, is in the v2.1.88 release that landed this morning alongside the source map controversy.
The cache mechanics compound the problem structurally: the default cache lifetime is just five minutes, which means even a short work break can flush the cache and inflate costs significantly on resumption. Upgrading to a one-hour cache lifetime doubles write token costs. The three-cause explanation now provides developers with the first complete accounting of why quotas have felt broken for the past week — and points to the v2.1.88 upgrade as the path forward, once the dust from today's packaging incident settles.