Anthropic's AI chatbot Claude has introduced usage limits, but users report hitting the cap quickly even under typical use.

On April 14, online media outlet Gigazine reported that a user on the $100-a-month Max 5x plan raised the issue on GitHub, saying the allowance was exhausted in 1 hour 30 minutes, making it effectively difficult to use for work.

Claude manages usage in 5-hour blocks. Once usage reaches a certain level within a block, additional use is restricted until 5 hours have passed from the start of the first use. The plan cited is a product with a usage cap 5 times higher than the $20-a-month Pro plan.

The user said they carried out intensive development work from 3 p.m. to 8 p.m. on the day of the issue. During the 5 hours, there were 2,715 API calls, and the maximum context length rose to about 970,000 tokens. The automatic context summarisation function also ran twice. The user said they could accept being limited after that level of heavy use.

The issue was what came after. The user said that even after 8 p.m., they used Claude in a typical way, such as light development work and questions and answers, but reached the usage limit again within 1 hour 30 minutes. The user said that while analysing the cause, they found indications that a Claude session left open in the background performed large amounts of cache reading.

The user suggested that while cache inputs are treated as one-tenth of normal inputs in cost calculations, the usage limit may not have reflected that. The claim was that in the limit calculation, cache inputs may have been counted at their original amount rather than at one-tenth.

The complaint also ties into Claude's large-context strategy. Claude Code provides paying users with a context window of up to 1 million tokens. That was presented as a strength because it can process more information at once, but the user pointed out that if cache reads are counted at full speed, a wider context window could increase input tokens per API call and make it easier to hit usage limits. In other words, the user argued that support for 1 million tokens could accelerate the depletion of the cap rather than being an advantage.

The user also argued that an idle session that is merely open in the background, without user interaction, should not use a large amount of APIs. That is because if session activity unrelated to actual work encourages cap depletion, the perceived value of the plan could fall sharply.

The Claude Code development team also moved to respond. One staff member said the team would review measures such as narrowing the default context window further and being more proactive in cleaning up background tasks. The team did not acknowledge the usage calculation structure itself, but it has taken steps to reduce the issue.

The issue also connects to how Anthropic operates its service. Anthropic is growing quickly, with revenue rising more than threefold over the past 3 months, but it has also been discussed as being short of computing resources in the short term. The company has previously said it would make usage limits stricter to handle load.

As a result, users have responded that Claude's response quality has recently dropped noticeably, while others have said the cache validity period has shortened, increasing usage consumption. The dispute appears to again show the gap between computing resource constraints, plan design and perceived usability for high-performance AI services.