Anthropic appears to be A/B testing reduced effort levels in Claude Code

A developer on Hacker News claims Anthropic is running a server-side A/B test on Claude Code 2.1.236+ that shrinks the effort scale. The test affects roughly 5% of sessions, while older versions and Opus 5 are left unchanged, and the changelog is silent about it.

Specifically, since version 2.1.237, the model interprets 'high' effort as 10 out of 100—the exact value 'low' used to have. This makes models seem noticeably less capable, and the user spent an afternoon convinced their own tools and app were broken before realizing the cause.

The post highlights a lack of transparency around Anthropic's product experiments, which can confuse developers who rely on consistent behavior from Claude Code.