When you signed up for the Max 5x plan at €100, you didn’t just buy intelligence; you bought a static expectation of capacity. In the early days, “usage limits” felt like a solid wall: you knew where they were, and when you hit them, the rules of engagement were clear. But mid-contract, Anthropic shifted the very architecture of your access into something far more predatory: The Rolling 5-Hour Window.
The Mechanics of the Cheat
The “rolling” nature of this window is diabolical because it creates a perpetual state of scarcity. In a static window (the one you paid for), if you had a 50-message limit for the morning, you’d use them and wait for the reset. In the Rolling Window, every single message you send is tied to its own independent 5-hour expiration timer.
Imagine a bucket with a hole in the bottom. You pour in “water” (messages). The hole only lets water out exactly 5 hours after it was poured in. If you pour a gallon in at 9:00 AM, that gallon stays there, taking up space, until 2:00 PM. If you pour in more at 10:00 AM, your bucket overflows (limit reached) even though it’s been an hour. You are trapped by your past self’s intensity.
Why It’s “Diabolical” (and how they cheat)
The Invisible Shrinking Bucket: Anthropic recently admitted to “adjusting” limits during peak hours (5 AM – 11 AM PT). Because they never tell you the exact number of messages in your 5x bucket, they can shrink the bucket size during your workday without “breaking” their terms. Your “5x” multiplier remains, but it’s 5x a much smaller number.
The Cache Tax & The 5-Minute Trap: While Anthropic claims Max users enjoy a 1-hour cache Time-to-Live (TTL) for the main agent, the reality—especially for Claude Code (CLI) users—is a minefield. The API default and any spawned subagents operate on a strict 5-minute fuse. Worse, a confirmed March 2026 regression silently forces even the main agent into this 5-minute limit under heavy load. Combine this with the “Resume Trap” (where resuming a closed terminal session instantly busts the cache) and the new 1M token context window, and the results are catastrophic. If you take a 6-minute coffee break with a large codebase loaded, Claude “forgets” everything. Your next reply forces a full re-read. Because your quota is tied to compute cost, this single cache miss can devour 5-10% of your entire 5-hour allowance in one keystroke, taxing you as if you’d just sent 50 new messages.
The Black Box: By refusing to provide a visualization tool, Anthropic forces you to play a mental game of “Quota Tetris.” You are left guessing when your 9:00 AM “burst” will finally evaporate so you can ask one more question.