If you are paying $100 a month for an AI coding tool and running out of tokens before free users do, something is wrong. That is the situation Anthropic is now publicly acknowledging.
The company posted on Reddit that it is investigating an issue where Claude Code users are hitting usage limits faster than expected, calling the fix the "top priority" for the team. The post confirmed what many developers had already been experiencing - token budgets evaporating at rates that do not match what customers reasonably expect when they pay for a service.
What Users Are Reporting
The developer feedback under the Reddit post paints a clear picture. One user reported hitting their token limit significantly later on their free account than on their $100 per month paid plan - the opposite of what any paying customer should experience. Another described a debugging loop where a single session consumed an entire daily budget in minutes. A third noted the problem extends beyond Claude Code entirely, describing a single one-sentence reply to a conversation that moved their usage from 59% to 100% with no warning and no obvious explanation.
The core frustration is opacity. Customers buy tokens to use AI services but the amount consumed per task is not always transparent, making it nearly impossible to manage usage proactively or predict costs reliably.
The Context Makes It More Complicated
The timing is awkward for Anthropic. Just last week, the company introduced peak-hour throttling on Claude, meaning tokens are consumed more quickly when demand for the service is higher. That decision now looks directly connected to the complaints surfacing from Claude Code users - paying customers who thought they were buying a fixed resource are discovering that resource depletes faster depending on when they use it.
Claude Code has also been in the news for other reasons. Anthropic recently acknowledged that an internal file containing 500,000 lines of its Claude Code source code was accidentally released on GitHub due to human error. The company said no sensitive customer data or credentials were exposed. It was not the first time - an earlier version had been leaked in February 2025, and independent developers had already partially reverse-engineered the codebase.
Why This Matters for Business Users
For software developers using Claude Code as part of their daily workflow, unpredictable token burn is not a minor inconvenience - it disrupts work mid-task and forces users to either stop entirely or upgrade to a higher pricing tier to continue. Claude Pro starts at $20 per month, with tiers reaching $100 and $200 for higher usage. When those higher tiers behave unpredictably, the value proposition breaks down.
I have seen this pattern before in enterprise software. When pricing and usage transparency do not match customer expectations, it does not matter how good the underlying product is. Trust erodes fast. Anthropic's decision to acknowledge the issue publicly and name it as a top priority is the right call. Execution on the fix is what matters next.
