SemiAnalysis: Anthropic is A/B-testing new usage limits on some accounts — ~20% less monthly total, ~20% more 5-hour allowance

SemiAnalysis reports that Anthropic is A/B-testing Claude subscription limits: test-group accounts get roughly 20% higher 5-hour allowances but about 20% less total monthly usage (measured in API value). Many users on X have been complaining that their Claude limits shrank, and SemiAnalysis says they are not wrong — only a subset of accounts is affected. Usage limits directly shape subscriber experience and effective cost, so the experiment reveals Anthropic's trade-off between inference costs and user retention, and may foreshadow changes to its subscription tier structure.

• 宝玉 (Baoyu)2026-10-23 00:19