Anthropic says Claude Code did get worse — but shoots down speculation it 'nerfed' the model

Anthropic has acknowledged user feedback regarding a decline in the performance of Claude Code, one of its key offerings. In a detailed blog post, the company confirmed it had identified three specific issues that likely contributed to a perceived decrease in quality. The announcement comes after a wave of complaints from users expressing dissatisfaction with the tool's recent performance. The company emphasized that it does not intentionally degrade its models, stating, "We take reports about degradation very seriously." Anthropic clarified that the underlying model itself remained unaffected, attributing the issues to modifications made at the product level. As of April 20, these problems have been addressed, and measures have been implemented to prevent future occurrences. The response follows a series of successes for Anthropic, which recently achieved a valuation of $1 trillion in secondary markets, thanks to the technical prowess of its tools. However, user frustrations peaked when some speculated that the company had deliberately 'nerfed' the tool, a claim Anthropic firmly rejected. Prominent users, including Stella Laurenzo from AMD's AI division, voiced their concerns about the tool's reliability. In the wake of these complaints, Anthropic's developers took to social media to summarize their findings and actions. They identified three main changes that led to the reported issues: an adjustment to Claude Code's default thinking level, a bug introduced by a cache-optimization tweak, and a modification aimed at reducing verbosity in responses. To rectify these issues, Anthropic has increased the involvement of its staff with the public build of Claude Code, enhanced its code review processes, and tightened controls around system prompt adjustments. In a gesture of goodwill, the company announced it would reset usage limits for all subscribers effective Thursday, thanking users for their valuable feedback. Following the announcement, many users expressed relief and validation. One user on the platform X remarked, "I can't believe we were right!" referring to a previous observation about the tool's performance dip coinciding with a change made on March 4. In addition, Anthropic faced scrutiny earlier this week regarding an experiment that temporarily removed Claude Code from its Pro plan for a small percentage of new users, which the company clarified was a limited test.