Anthropic fixes Claude Code bugs v2.1.116+

Anthropic said three separate changes degraded Claude Code over the past month, and all three were fixed by April 20 in version 2.1.116. (anthropic.com) The company said the problems affected Claude Code, the Claude Agent SDK, and Claude Cowork, but not the Claude API or Anthropic’s inference layer. (anthropic.com) The first change landed March 4, when Anthropic lowered Claude Code’s default reasoning effort from high to medium to cut long waits that sometimes made the interface look frozen. Anthropic said it reverted that decision on April 7 after users said they preferred higher intelligence by default. (anthropic.com) (github.com) The second problem arrived March 26, when a feature meant to clear older thinking after an hour of session idleness kept clearing it on every turn instead. Anthropic said that made Claude seem forgetful and repetitive until the bug was fixed on April 10. (anthropic.com) The third change came April 16, when Anthropic added a system prompt instruction to reduce verbosity. The company said that instruction, combined with other prompt changes, hurt coding quality and was reverted on April 20. (anthropic.com) Anthropic said the three issues hit different slices of traffic on different schedules, which made the degradation look broad but inconsistent. The company said that pattern also made the reports harder to reproduce in its internal usage and evaluations. (anthropic.com) The affected model lineup changed by bug. Anthropic said the reasoning and session issues hit Claude Sonnet 4.6 and Claude Opus 4.6, while the verbosity prompt also affected Claude Opus 4.7. (anthropic.com) Anthropic said on April 23 that it was resetting usage limits for all subscribers after the fixes. The Claude Code changelog also shows the default effort for Pro and Max subscribers on Opus 4.6 and Sonnet 4.6 is now high again, not medium. (anthropic.com) (github.com) Claude Code is Anthropic’s coding agent for terminal-based software work, and the post-mortem amounts to an acknowledgment that recent quality complaints reflected product changes, not imagined drift. Anthropic said it “never intentionally degrade[s]” its models and that the failures came from these specific changes instead. (anthropic.com)