SemiAnalysis Reports Anthropic is Testing Quota Adjustments for Some Claude Accounts
SemiAnalysis has detected that Anthropic is conducting A/B testing on the subscription quotas for Claude, affecting only a portion of accounts.
The rolling quota for test group accounts has increased by approximately 20% for a 5-hour period, but the total monthly amount has decreased by about 20% based on API equivalent pricing.
Recent user reports on social media about reduced Claude quotas are partially correct, as affected accounts have experienced a contraction in monthly totals.
SemiAnalysis concluded this through continuous tracking of changes in subscription usage limits, having previously measured that a very small-scale A/B test by the same provider had caused individual account quotas to drop by about 20%.
The API equivalent value provided by Anthropic's subscription plan for mid-tier models is significantly higher than competitors, but quotas can be unilaterally modified by the provider without publicly adjusting prices.
Subscription revenue constitutes a low proportion of Anthropic's total revenue but consumes a large amount of inference computing power, and quota adjustments affect the actual user experience for heavy and light users.
In terms of funding and computing power allocation, increasing short-term quotas may encourage concentrated usage, while reducing the monthly total compresses the overall subsidy intensity. The test aims to optimize user behavior and cost structure.
Source: Public Information
ABAB AI Insight
SemiAnalysis has long tracked the API equivalent value of subscription plans from companies like Anthropic and OpenAI through actual usage metrics. It previously found that the mid-tier model plan for Claude offers about 5 times the equivalent usage compared to OpenAI and recorded that OpenAI had halved the quotas for some of its high-priced plans. Anthropic's official website reserves the right to adjust weekly and monthly limits at any time.
In terms of capital pathways, the subscription plan exchanges a fixed monthly fee for high quota subsidies but consumes disproportionate inference computing power. By adjusting the 5-hour and monthly ratios through A/B testing, it can redistribute usage windows without changing public prices, reducing heavy users' occupancy of fixed resources.
Similar cases include dynamic adjustments of API rate limits by cloud computing vendors and SaaS companies guiding user behavior through usage tiering. The current phase is transitioning from high-subsidy expansion of large model subscriptions to optimization of costs and usage efficiency.
Essentially, this represents a transfer of pricing power and restructuring of the industry chain: quota adjustments under fixed prices serve as implicit pricing tools, shifting subsidies from total amounts to time windows, allowing providers to optimize computing power allocation and user stickiness while controlling overall inference costs.
ABAB News · Cognitive Laws
- Quota adjustments under fixed prices are implicit pricing tools
- Short-term quota increases are often accompanied by total reductions
- Subsidy intensity determines user behavior rather than public prices.