Claude Admits to Excessive Charges

After a wave of complaints from users about unexpected billing issues with Claude Code, Anthropic has finally responded, acknowledging that users are reaching their usage limits much faster than anticipated. The team is currently investigating the issue as a top priority and will provide updates soon.

The situation escalated when a user reverse-engineered Claude Code and discovered two independent bugs, which were shared on Reddit. These bugs reportedly cause the prompt cache to fail, leading to inflated billing costs that could increase by 10 to 20 times without user awareness.

Insufficient Packages

Tokens are often compared to utilities in the modern era, but using them feels more like early phone bills or data plans:

It seems like you’re not using much, yet it’s never enough.

One user noted that a simple greeting could consume 13% of their token limit.

In just 11 minutes of work, 23% of their usage was gone.

In a shocking case, a single prompt consumed 31% of the total limit.

Even the highest-tier plan at $200 per month reached its limit in just three and a half hours.

A user subscribed to Claude Pro (annual fee of $200) complained on Discord that they often hit their limit by Monday and had to wait until Saturday for a reset, effectively only using the service for about 12 days each month. This rapid consumption has rendered Claude Code nearly unusable for those who rely on AI for work.

Systemic Issues

The complaints are not isolated; many users have reported that major AI services have similar issues, often exhausting their monthly limits in the first three weeks without heavy usage. However, Claude Code’s recent issues seem particularly egregious, leading to widespread dissatisfaction on Reddit.

The Register suggests three potential reasons for these problems:

1. Last week, Anthropic announced that they would reduce quotas during peak usage times, meaning users would experience faster depletion of their limits during busy hours (08:00–14:00 ET on weekdays).

2. March 28 was the last day of a promotional event that doubled usage limits during non-peak hours. With the promotion over, users returned to normal limits, which felt significantly reduced.

3. The two bugs discovered by Reddit users: one affects the sentinel replacement mechanism in the billing logic, while the other causes the cache to fail, leading to repeated calculations of the same requests. This results in token consumption being inadvertently amplified, potentially increasing costs by over ten times.

Some users reported that downgrading to an older version of the software improved their experience significantly.

Sustained Usability Over Model Capability

Ultimately, the issue with Claude isn’t just about the model’s strength but whether users can reliably access it. As model capabilities improve, the demand for stable usage increases, making it crucial for service providers to respond quickly to user feedback and maintain consistent delivery.

While users may be willing to pay for high-quality responses, the rising costs and deteriorating experience have led to frustration. Users have pointed out that even basic token management seems to be poorly handled by Anthropic’s support.

Despite continuous feedback over several days, Anthropic’s response has been slow, only acknowledging the issue after users identified the bugs.

In contrast, competitors like OpenClaw have been more responsive, frequently releasing updates and fixes.

This raises a critical question in the AI era: the scarcity may not be in model capability but in the ability to deliver consistent service and respond to user needs effectively.

If Claude’s code is AI-generated, perhaps Anthropic should consider hiring more staff for customer support.