One prompt. That's all it took for one Claude Max 20x subscriber ($200/month) to watch their usage jump from 21% to 100%. They're not alone.
Starting around March 22-23, Claude Code users across Pro, Max 5x, and Max 20x plans began reporting that their 5-hour rolling session limits were draining dramatically faster than before. Sessions that used to last the full window started hitting walls in 1-2 hours. Individual prompts that previously consumed modest chunks of quota began eating 20-50% in a single exchange.
The reports are specific and consistent. A Max 5x subscriber ($100/month) hit 58% usage in 40 minutes. Another user saw 3% of their quota vanish from a single "hey Claude" message. Multiple users documented their usage counters climbing even after terminating all Claude Code sessions, revoking API keys, and logging out of every browser.
What Anthropic Said
On March 27, Anthropic acknowledged adjusting session limits during peak hours (weekdays 5am-11am PT). The company said users would "move through limits faster" during those windows and estimated about 7% of users would hit limits they previously wouldn't have.
That explanation hasn't satisfied the community, and for reasonable cause. Users reported the same issues outside peak hours. The idle counter increment bug - where usage climbs with no active sessions - doesn't fit a peak-hours adjustment. And the /usage command in Claude Code sometimes showed 100% (blocked) while claude.ai settings displayed only 73%, suggesting a sync problem between tracking systems.
The GitHub Response Made It Worse
Multiple detailed bug reports filed on GitHub (issues #37394, #38335, #39649) were labeled "invalid" or auto-closed as duplicates without visible responses from Anthropic engineers. For users paying $100-200/month who couldn't use the product they were paying for, the silence felt worse than the bug.
No refunds or credits have been publicly offered.
The Practical Fallout
The timing is particularly bad. Claude Code has been gaining serious traction among developers who use it as their primary coding tool. Reliability of access matters more than raw capability when a tool is embedded in your daily workflow. Some users have started shifting work to off-peak hours as a workaround, which helps marginally but doesn't address the underlying anomalies.
Others are hedging their bets. The most common suggestion in community discussions: use Google's Gemini CLI for broad codebase exploration (where its generous free tier shines), then switch to Claude Code for the work that requires sharper reasoning. It's a pragmatic response, but not one Anthropic wants to encourage.
If you're a Claude Code subscriber experiencing this, the /usage command shows your current consumption. Shifting heavy sessions outside weekday mornings PT may help. But the real fix needs to come from Anthropic's engineering team, and so far, they haven't confirmed whether this is a bug or an intentional change that went further than planned.