Comparing changed usage limits for some users?
U.S. semiconductor research firm SemiAnalysis published an analysis on X and in a members-only article on Oct. 22 saying Anthropic is running a test (an A/B test) that changes usage limits on Claude's paid plans for only some users and compares the results. On the 23rd, X user @notjazii shared the analysis, and it spread.
According to the analysis, the usage limit for each five-hour window increases by about 20%. On the other hand, the amount usable in a month, converted to API pricing, decreases by about 20%. The monthly fee users pay does not change. SemiAnalysis said it applies to "only some accounts" and did not say which plans or how many users are affected.
Monthly usage can fall even as the 5-hour limit rises
Claude's paid plans have a limit that resets every five hours and a weekly cap. Because the five-hour limit resets many times, if the weekly cap becomes stricter, the amount usable in a month falls even if the five-hour limit increases. SemiAnalysis said it made its estimate by observing the usage meter while assuming a fixed set of tasks and separating consumption by type of token. The estimate varies depending on the assumed usage.
Reports of routing to a new model
@notjazii's post also relayed reports that some requests to Sonnet 5.5 are being routed to a new model. ORIENS has not been able to confirm this routing. Anthropic has not announced anything about either the usage limit test or the routing.
Usage limits have become a factor in how AI companies compete on the value of their paid plans. OpenAI introduced a $500-a-month tier and, from Oct. 30, halved the usage limits for tasks on its $200-a-month Pro plan. SemiAnalysis continues to analyze and compare companies' usage limits by converting them to API pricing.
