Anthropic has begun quietly reshaping how customers access its Claude models, introducing a new system that effectively reduces available computing power during peak hours while preserving overall weekly usage limits.
The change, disclosed in a social media post by technical staff member Thariq Shihipar, comes amid growing pressure on the company’s infrastructure as demand for generative AI tools continues to surge.
“To manage growing demand for Claude we’re adjusting our five hour session limits for free/Pro/Max subs during peak hours. Your weekly limits remain unchanged,” Shihipar wrote.
Register for Tekedia Mini-MBA edition 20 (June 8 – Sept 5, 2026).
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Register for Tekedia AI Lab.
In practical terms, the adjustment alters how time is measured. Claude’s subscription tiers, ranging from free access to paid plans, operate on a “five-hour session” model. But that time is not fixed in real-world hours; it is tied to token consumption, a metric that reflects how much computational work a user’s prompts and outputs require.
Under the new regime, users operating during peak demand windows—defined as 05:00 to 11:00 Pacific Time (13:00 to 19:00 GMT)—may exhaust what is nominally a five-hour session in significantly less time if their workloads are intensive. Outside those hours, the same allocation stretches further, effectively delivering more usable compute for the same subscription.
The company has not disclosed the exact token thresholds behind these limits, maintaining a long-standing opacity around how usage is calculated. That lack of transparency has been a recurring point of friction for developers and power users, who often struggle to predict how quickly their allowances will be consumed.
Shihipar acknowledged the uneven impact. ~7 percent of users will hit session limits they wouldn’t have before, particularly for pro tiers. If you run token-intensive background jobs, shifting them to off-peak hours will stretch your session limits further,” he said.
Anthropic says the changes are neutral over a full week. Capacity has been expanded during off-peak periods, allowing users to recover lost ground if they adjust their usage patterns.
“Overall weekly limits stay the same, just how they’re distributed across the week is changing,” Shihipar added. “I know this was frustrating. We’re continuing to invest in scaling efficiently. I’ll keep you posted on progress.”
Anthropic is the only AI company facing this challenge, which underlines a broader infrastructure issue in the industry. Demand for large language models is rising faster than the infrastructure needed to support them. Training and running advanced models require vast computing resources, and even well-funded firms are being forced to ration access during periods of heavy use.
Anthropic offers its services through both an application programming interface, where customers pay per token, and subscription plans with bundled usage. While API pricing is transparent, covering input tokens, output tokens, and various caching mechanisms, subscription limits remain less clearly defined, governed by internal formulas that factor in conversation length, model choice, and feature usage.
“Your usage is affected by several factors, including the length and complexity of your conversations, the features you use, and which Claude model you’re chatting with,” the company notes in its documentation. “Different subscription plans (Pro, Max, Team, etc.) have different usage allowances, with paid plans offering higher limits.”
For developers and enterprise users, the implications are operational. Workloads that can be scheduled, such as batch processing or background tasks, will increasingly be pushed into off-peak windows to maximize efficiency. Real-time use during peak hours, by contrast, becomes more expensive in terms of consumed allowance, even if pricing remains unchanged.
The adjustment also underscores a shift in how AI services are being delivered. Rather than offering fixed access, providers are moving toward dynamic allocation models that mirror cloud computing—where capacity, performance, and availability fluctuate based on system load.
That means user access is no longer just a function of subscription tier, but of timing and workload intensity. Anthropic sees it as a way to stretch limited resources without formally raising prices or imposing stricter caps. However, the trade-off is predictability. As demand continues to climb, managing when and how to use AI tools is becoming as important as deciding which tools to use in the first place.