Anthropic Explains Claude Code Performance Degradation

Anthropic acknowledged three engineering changes that caused the Claude Code performance degradation.

• March 4: The default reasoning effort was lowered from high to medium to reduce latency, but this worsened code quality.
• March 26: A bug was introduced that kept discarding the model's reasoning history during sessions, making it appear more forgetful and unstable, and also caused usage limits to be consumed faster.
• April 16: A system prompt limiting responses between tool calls to 25 words was added, degrading coding quality; this measure was reverted 4 days later.

Anthropic stated that all these issues were resolved as of April 20, and that the API was not affected. It then reset usage limits for all subscribers on April 23.

However, user backlash had already grown significantly. Some canceled their subscriptions, and suspicions spread that the company was hiding a compute shortage. Anthropic explained that surging demand had put pressure on its infrastructure, especially during peak hours, and that it is securing additional capacity through expanded partnerships with Amazon and Google.

This controversy could also weigh on developer loyalty to Claude Code, Anthropic's pricing policy, and its competition with OpenAI over developer tools.