Anthropic tweaks timed usage limits to discourage Claude demand during peak hours

In the world of generative artificial intelligence, where computing power has become the new currency, market leaders must seek creative ways to manage infrastructure. Anthropic, the creator of the acclaimed Claude model, has just introduced unprecedented changes to how usage limits are calculated. As of Wednesday, the company is officially modifying the rules for accessing its services, introducing a mechanism that could be tentatively called a "peak tariff." In practice, this means that an hour of work with AI at noon may cost the user significantly more of their limit than the same hour spent in a session in the middle of the night.

This decision is a direct response to rapidly growing demand, which is beginning to test the performance limits of the company's server rooms. Instead of a simple subscription increase or a drastic cut in functionality, Anthropic has opted for a dynamic model designed to discourage intensive use of the Claude model during global peak hours. This is a signal to the industry: free and unlimited AI resources are definitively a thing of the past, and operational cost optimization is moving down to the level of the individual user query.

Peak hours in the computing cloud

The new system operates on specific time windows that Anthropic has defined as periods of increased traffic. These are the hours between 05:00 and 11:00 PT (which corresponds to 13:00 – 19:00 GMT). During this time, users of Free, Pro, and Max subscriptions may find that their five-hour session limits are exhausted much faster than the clock would suggest. In practice, a five-hour time allocation can be "burned" in a much shorter period if interactions with the model are intense.

This mechanism is possible because Anthropic ties time limits to actual token consumption, although the methodology of these calculations remains undisclosed. While outside peak hours five hours of access actually means 300 minutes of work, during peak times the algorithm becomes more restrictive. Thariq Shihipar from the company's technical team admitted that approximately 7 percent of users – particularly those using Pro plans – will feel these changes through earlier session blocks that they would not have experienced before.

Token mathematics and hidden limits

For the average user, the limit system in Claude remains quite opaque. Unlike customers using the API, who pay according to a clear price list for Base Input Tokens, Cache Writes, or Output Tokens, subscription plan subscribers operate in a sphere of estimates. Anthropic currently offers four main subscription levels:

• Free – basic access with the lowest priorities.
• Pro – costing $20 per month.
• Max 5x – priced at $100 per month.
• Max 20x – the highest plan at $200 per month.

However, even the highest plans do not guarantee complete immunity to the new regulations. The company explains in its documentation that limit consumption depends on the length and complexity of the conversation, the features used, and the specific Claude model being interacted with. What is new is that a real-time variable is now added to this equation. Although users have access to a dashboard showing progress in session limit consumption and weekly allocation, they lack the tools to precisely plan their token efficiency before starting work.

Load shifting strategy

The move taken by Anthropic is a classic example of "load balancing" transferred to the ground of end-customer relations. The company openly encourages developers and business users to shift heavy tasks, such as processing large datasets or background jobs, to night or morning hours outside the peak window. This is intended to allow for "stretching" available session limits without the need to pay extra for additional packages.

It is worth noting that Anthropic has simultaneously increased performance during low-traffic hours. According to Shihipar's assurances, overall weekly limits remain unchanged – only their distribution is being modified. This is an attempt to mitigate the frustration of users who, paying 200 dollars a month, expect reliability regardless of the time of day. However, the company argues that this is a necessary step to maintain service stability for all subscribers in the face of exponential growth in interest in Claude models.

It can be expected that the "dynamic session cost" model will soon become a standard across the entire AI industry. As models become increasingly resource-intensive and access to the latest GPUs remains a bottleneck, service providers will be forced into increasingly aggressive demand management. Anthropic is one of the first to openly admit that time spent with a model is not equal to CPU time, and users must learn to optimize their interactions with AI not only for prompt quality but also for the clock on the wall.