User complaints about Claude’s Token consumption and quality issues triggered heated community discussions: Crisis of trust in AI tools

TL;DR

Developer Nicky Reinert published a long article on his personal blog, recording the entire process of canceling his Claude Pro subscription. The article points out three core problems: Token quota is consumed abnormally during off-peak hours, model answer quality continues to decline, AI customer service is unable to solve actual problems, and manual support response formatting is available.

• Time of incident: April 24 (blog post release time)
• Discussion platform: Hacker News homepage, 816 points and 488 comments
• Core controversial points: Token consumption mechanism is opaque, model quality is degraded, and support experience is poor

Text

User cancellation Claude: Three core complaints trigger industry reflection

On April 24, 2026, developer Nicky Reinert published an article titled "Why I Cancelled Claude: Token Issues, Declining Quality, and Poor Support" on his personal blog, recording in detail the complete process of canceling his Claude Pro subscription. This article quickly appeared on the home page of Hacker News, receiving more than 800 likes and nearly 500 comments, becoming one of the most popular discussion topics in the AI ​​field that day.

Reinert's complaints focus on three areas. The first is the token consumption problem: he only asked two small questions when he started working in the morning, and the token quota was quickly exhausted to 100%. The second is that the quality of answers continues to decline: initially he was able to work on multiple projects at the same time, but later he could exhaust the token limit by completing a single project within two hours, and the output quality was significantly lower than the initial experience. The third is poor customer service experience: the AI ​​support robot only returned a general template reply, and subsequent human customer service did not provide solutions to specific problems.

Community reaction: This is not an isolated case

Reactions in the Hacker News comments section suggest that Reinert's experience is not unique. Many users shared similar experiences, mainly focusing on the negative impact on user experience after Anthropic’s recent adjustments to the token mechanism. Some comments point out that Anthropic seems to have degraded the resource allocation quality of older models while launching new models, such as Claude Opus 4.7.

Some users mentioned in comments that the subscription-based business model in the AI ​​industry is leading to a "boiling frog in warm water" decline in service quality - the initial experience is excellent to attract users to subscribe, but after subscription, resource allocation and service quality are gradually reduced.

Impact Analysis

Industry reaction

Hacker News user pointed out: "This is not a problem unique to Claude. The entire AI industry faces a balance between growth pressure and user experience. When companies prioritize serving enterprise customers and pursuing revenue growth, individual users are often the first to feel the decline in service quality."

Another commenter mentioned that similar token consumption issues have also appeared on OpenAI and Google's platforms. The root cause is that the cost of AI model inference remains high, and suppliers balance profit margins by blurring token usage strategies.

Adaptation suggestions

For developers using AI tools, there are several points worth paying attention to in this case:

• Build a multi-tool backup: Don’t completely tie your workflow to a single AI platform
• Monitor Token Consumption: Regularly check whether there is abnormal growth in usage
• Focus on alternatives: DeepSeek, OpenAI, Google and other platforms are competing for the developer market, and competition brings better prices and experiences

*This article is news information for reference only and does not constitute any investment advice. The data sources are public reports and community discussions. *