Anthropic's Claude Code performance decline triggers strong user dissatisfaction

ME News Report, April 15 (UTC+8), Anthropic is facing criticism from some core users due to a decline in the performance of its Claude model. These issues have led the company to deal with external speculation about its motives and service capabilities as it reportedly prepares for an IPO. Many developers and heavy users have stated that the performance of Anthropic’s Claude model has recently significantly declined, characterized by difficulty in following instructions, choosing inappropriate shortcuts in complex tasks, and an increase in error rates. User complaints are mostly concentrated on Anthropic’s programming tool Claude Code. The problem seems related to recent adjustments Anthropic made to how the model runs, where the company lowered the default “effort level” to reduce the number of tokens processed per request, thereby saving computational resources. There is speculation that after a recent surge in product usage, Anthropic may be facing a shortage of computing resources. Anthropic has not responded in detail to the related complaints. Boris Cherny, head of the Claude Code product, responded to user complaints online, stating that the company has set the default “effort level” to “medium” in response to previous feedback about high token consumption, but some users believe the company did not fully disclose this change. (Jin10) (Source: BlockBeats)