Claude was supposed to be the careful, precise alternative - the one that didn't hallucinate details or water down answers. A developer writing at nickyreinert.de just published a detailed account of why that stopped being true for them, and cancelled their Claude Pro subscription over it.
The three complaints in the post will sound familiar to anyone who's spent serious time with Claude: token limits that interrupt work mid-task, output quality that seems to have declined, and customer support that offered nothing useful in response.
The token issue is the most concrete. Tokens are the units of text the model processes at once - roughly 4 characters each, so 100,000 tokens covers about 75,000 words. When a session fills the available window, conversations truncate or have to restart, which breaks long coding sessions, document reviews, or multi-step research tasks. Pro subscribers expect those limits to be high enough not to matter. For some users, they aren't.
The Quality Regression Problem
The quality decline complaint is harder to nail down but shows up often enough to take seriously. Users describe responses that felt detailed and technically sharp a few months ago now coming across as generic, hedged, or less accurate. Whether that reflects actual changes to the model, reduced compute per query during busy periods, or just expectations rising faster than improvements - nobody outside Anthropic can say for certain.
What's notable is that Anthropic hasn't addressed quality regression complaints publicly. When a paid product starts feeling worse and the company says nothing, the silence itself becomes a reason to cancel.
The support failure makes it worse. AI tool support is inconsistent across the industry, but for a $20/month product that professionals use as a core part of their work, "support didn't help" isn't a minor complaint - it's the thing that pushes someone from frustrated to gone.
What One Cancellation Actually Signals
A single blog post isn't a trend. But the specific combination - token constraints, quality inconsistency, unresponsive support - comes up constantly in user discussions. These aren't niche gripes from power users with unusual workflows. They're complaints from the exact professionals Claude is supposed to retain.
Claude's value proposition rests almost entirely on being more reliable and precise than alternatives. If that perception erodes, the subscription math changes fast. ChatGPT and others are always one cancellation decision away from picking up those users.