Anthropic is modifying how fast users burn through their Claude usage limits during busy periods. To manage growing demand and computing strain, the artificial intelligence company is tightening session caps for free and paid subscribers during peak weekday hours.
This unannounced adjustment arrives shortly after the company ran a promotional event that temporarily doubled allowances for users during off-peak times. The platform has also experienced recent growing pains, including a major service disruption that lasted nearly five hours.
Tighter Restrictions During Peak Demand
The new restrictions affect individuals on the free tier, as well as those subscribed to the Pro and Max plans. During peak hours—defined as 5 a.m. to 11 a.m. Pacific Time or 1 p.m. to 7 p.m. Greenwich Mean Time on weekdays—users will deplete their five-hour session allowances much faster than before.
Unlike some competing AI platforms that use daily caps, Claude operates on a rolling five-hour window. Once a user hits their cap within that timeframe, they must switch to a less capable model or wait for the session to refresh. While the speed of reaching these limits has increased during busy times, weekly usage limits remain unchanged.
Thariq Shihipar, an engineer working on Claude, announced the changes on social media rather than through an official company channel. He noted that roughly seven percent of users will hit session limits they previously would not have encountered, especially those on professional tiers. To maximize session time, Shihipar advised users to move heavy, token-intensive background jobs to off-peak hours.
The company offers a pay-as-you-go system for paid subscribers who want to continue working after hitting their plan restrictions.
Recent Promotional Boost
The tightening of limits contrasts with a recent temporary promotion. Anthropic previously announced on its official social media account that it was doubling usage limits outside of peak hours as a token of gratitude to its users.
This increased access applied to the free tier, along with Pro, Max, and Team plans, but excluded Claude Enterprise accounts. On weekdays, the doubled capacity was available outside of peak hours, which were noted during the promotion as 12 p.m. to 6 p.m. Greenwich Mean Time. Meanwhile, weekend users enjoyed doubled limits throughout the entire day.
The promotion applied automatically. It covered the main website and mobile applications, as well as integrated tools like Cowork, Claude Code, Claude for Excel, and Claude for PowerPoint. However, free tier users remained restricted from features not included in their standard plan. According to conflicting reports, the promotional period was scheduled to run through either March 27 or March 28 before limits reverted to normal.
Outages and Compute Strain
These usage adjustments come as Anthropic faces significant compute strain and technical difficulties. The platform recently suffered a major outage that lasted nearly five hours, marking the second disruption in a single week.
User complaints surged on tracking websites, with the Claude Chat interface accounting for half of the reported issues. The mobile application and Claude Code were also heavily impacted. While models like Opus 4.6 and Sonnet 4.6 experienced high error rates, Claude for Government services remained largely unaffected. Users trying to access the platform encountered error messages stating that operations were taking longer than usual.
The company is dealing with a massive surge in mainstream interest. This spike in attention followed reports that the Pentagon effectively blacklisted Anthropic after CEO Dario Amodei refused to provide the military with unrestricted access to the company’s AI models. Amodei has maintained that the company prefers to focus on its enterprise business operations.
Additionally, the rise of open-source AI agents, such as OpenClaw, has allowed users to leverage artificial intelligence models more intensively. This explosion of agent-based workflows means users are consuming computing power at unprecedented rates. The industry-wide challenge of balancing available compute has even forced competitors to change strategies, with OpenAI recently dropping its video generation tool to refocus resources on core services. In December, Amodei commented on the aggressive spending plans of competitors building massive data centers, suggesting that some industry players are taking unwise risks to meet future compute demands.
