Complaints about Claude's declining code quality have been building for weeks. We reached out last week, and Anthropic has now confirmed technical bugs and opened a deeper investigation.
Over the past few weeks, numerous developers and users on platforms like Reddit, X, and YouTube have reported a noticeable degradation in the performance of Anthropic's AI models, especially in code generation. The criticism ranged from general statements that Claude "has gotten much worse" to detailed bug reports. Now, Anthropic has officially responded to the allegations and acknowledged technical problems.
"Claude has lost its mind"
The complaints, which have been piling up mainly since late August 2025, focused on the model's programming abilities. A user on Reddit wrote that Claude had become "significantly dumber… ignored its own plan and messed up the code." Others reported that the model had started to "lie about the changes it made to code" or didn't even call the methods it was supposed to test. Our own development team also reported issues.
The community speculated about the causes. Some suspected deliberate "throttling the performance," while others considered technical reasons like the use of less reliable quantized variants of the model. There were also counterarguments that blamed the user, citing issues like Context Rot in long conversations or a growing complexity of their own codebase.
Anthropic admits bugs and promises improvement
When asked by THE DECODER last week about the numerous user complaints, Anthropic said in a statement: "We're aware of community reports about recent performance issues with Claude Code and take this feedback seriously. We recently resolved two bugs affecting Claude Sonnet 4 and Claude Haiku 3.5, and are investigating reports around output quality in Claude Opus 4.1."
The company emphasized the role of its users: "User feedback across multiple channels helped us identify and prioritize these incidents. We're thankful to those helping us make Claude better and are working to improve how we communicate with our developer community when issues arise."
These statements align with an official incident report in which Anthropic clarifies that it "never intentionally degrade[s] model quality as a result of demand or other factors." The issues were attributed to "unrelated bugs." The investigation into quality reports concerning the most powerful model, Claude Opus 4.1, is still ongoing. The problems affected Claude.ai, the developer console, the API, and Claude Code. The investigation into quality concerns with the most powerful model, Claude Opus 4.1, is still ongoing.
The incident brings back earlier debates: similar complaints surfaced in August 2024, and in the second half of 2023, users accused OpenAI of making GPT-4 "dumber" over time. OpenAI denied making major changes to the models after release, and many chalked up the frustration to changing user expectations - the initial "wow effect" wears off, and people start to notice the flaws. The Claude case now makes clear that technical problems can sometimes be behind these shifts. It also shows why more transparent communication from AI companies is crucial for keeping user trust, especially when model performance isn't always consistent.
