If your Claude session crashed right in the middle of a crucial morning workflow this week, you aren't alone. Battling surging developer demand, Anthropic quietly throttled capacity during peak weekday hours to keep its servers from melting under the load.

The New Peak-Hour Reality

Right now, Anthropic is aggressively tweaking its five-hour rolling usage limits for subscribers. If you're working between 8:00 a.m. and 2:00 p.m. Eastern Time on weekdays, your tokens drain at an accelerated pace. Consequently, users are hitting the dreaded session limit wall long before lunch.

Blame the shift toward heavy-duty agentic workflows and massive code generation tasks. Power users are churning through tokens faster than ever, prompting Anthropic to note that around 7 percent of customers—especially on Pro tiers—are suddenly hitting unfamiliar usage caps.

Total weekly compute limits haven't shrunk, though. Instead, the company simply reshuffled the deck on how daily capacity gets distributed. To stretch those session limits, subscribers running heavy background jobs will need to start pushing workloads to off-peak hours.

The Off-Peak Doubling Promotion

Trying to soften the blow of these midday restrictions, Anthropic rolled out a temporary capacity boost for night owls and weekend warriors. Between March 13 and yesterday, March 27, the company automatically doubled the message allowance within that rolling five-hour window.

This perk kicked in specifically during designated quiet periods. The bonus applied before 8:00 a.m. and after 2:00 p.m. Eastern Time on weekdays, plus around the clock on weekends.

Almost every subscription tier grabbed a piece of this massive capacity increase. Free, Pro, Max, and Team plan users saw their limits automatically jump across the web, mobile apps, Claude Code, and Microsoft Office integrations. Only Enterprise plan subscribers were left out of the promotion.

The strategy behind the off-peak incentive is simple: server load balancing. By bribing users with double the prompts and longer high-context sessions when it's quiet, the company cleverly pushes heavy compute tasks away from its most strained operational windows.

Infrastructure Under Pressure

Underneath these rate adjustments lies the brutal reality of infrastructure costs for modern frontier models. Since Anthropic rolled out its massive one-million token context window, user behavior has shifted drastically. Rather than just chatting, subscribers now routinely dump entire codebases into the prompt and lean heavily on Claude's "computer use" abilities.

To keep the lights on, the AI provider enforces a strict weekly "active hours" cap. Rather than watching the clock on the wall, an active hour ticks down only when Claude is actually processing tokens or reasoning.

Pro plans get roughly 40 to 80 active hours per week on Sonnet models. Meanwhile, Max tiers can access up to 480 Sonnet hours or 40 Opus hours, all depending on prompt complexity and session concurrency.

For developers and power users, the takeaway is clear: you have to optimize your scheduling just as much as your prompts. If your daily workflow relies on large-scale data analysis or automated coding, you'll need to schedule those heavy background jobs for your off-hours to avoid hitting a mid-morning dead end.