Anthropic announces Claude Code's reduced intelligence post-analysis: three layers of product modifications stacking, not a model issue

robot
Abstract generation in progress

CoinWorld News, the Anthropic engineering team announced that the decline in Claude Code quality stems from three independent product-level changes, affecting Claude Code, Claude Agent SDK, and Claude Cowork, with the API and underlying models unaffected.
The three issues were fixed on April 7th, 10th, and 20th, with the final version being v2.1.116.
The first change occurred on March 4th, when the team lowered the default inference strength of Claude Code from high to medium to reduce Opus 4.6 latency; after user feedback, it was rolled back on April 7th.
The second was a bug introduced on March 26th, causing the model to gradually lose inference context, resulting in forgetfulness and abnormal tool calls, fixed on April 10th.
The third change was deployed on April 16th, adding instructions to limit output length, which affected coding quality; it was rolled back on April 20th.
Anthropic stated they will require internal staff to use the same publicly built versions as users and conduct full model evaluations after each modification.
As compensation, all subscription users’ usage quotas have been reset.

View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments
  • Pin