SemiAnalysis: Anthropic A/B tests Claude subscription quotas, some users get 20% higher limits but lower value
Research firm SemiAnalysis reported that Anthropic is A/B testing Claude subscription quotas, giving some accounts on the same plan a 20% higher five-hour usage limit. However, when converted to API prices, these accounts' monthly resource value is about 20% lower. The report follows widespread user complaints about shrinking Claude quotas, and SemiAnalysis suggests quota settings can vary even for identical plans.
Coverage timeline
机器之心机器之心
编辑｜Neo、山辉 你敢信，花一样的钱订阅 Claude，你拿到的额度却可能比别人少了 20%。 根据研究机构 SemiAnalysis 的最新披露， Anthropic 正在对 Claude 订阅额度进行 A/B 测试，部分相同套餐的用户，被分配了不同的额度设置。 测试中，部分账号的五小时使用额度被提高了 20%。 听上去似乎是好事，但 SemiAnalysis 发现， 这部分账号按 API 价格折算，月度可用资源的价值反而降低了约 20% 。 怎么回事，难道真的「鱼越多鱼越少」？ Claude 额度越来越不够用了？ 最近，很多用户反映，Claude 的额度越来越不够用。 在 Reddit 的 ClaudeCode 社区，一位用户抱怨，自己原本可以用上四五天、甚至一周的额度，如今仅仅 24 小时就快要耗尽。 还有用户表示， 刚刚重置的额度，只进行几次高强度操作，就已经消耗了相当大一部分。 X 用户也有类似的遭遇，有人认为过去几周 Anthropic 一直在削减额度，甚至大胆预言了「500 美元」的新套餐。 实际上，Anthropic 近期确实频繁地调整 Claude 的额度，调整结果有增有减。 今年 5 月至 9 月，Anthropic 曾开展限时活动，将 Claude Code 的每周使用额度提高 50%。 但从 9 月 14 日起，这项活动结束，新的标准额度虽有提高，但换算下来，用户实际拿到的比活动期间减少了约 17%。 紧接着，9 月 22 日，Anthropic 发布 Claude Opus 5.5，又宣布提高 Pro、Max、Team 等套餐的 5 小时使用额度。按照 SemiAnalysis 当时的实测，使用 Opus 5.5 时，Max 套餐可用 token 数量增加约 20%，Pro 套餐则增加约 50%。 由于模型和推理强度不同，不同时期的相同订阅额度在使用体验上大不相同。同时，5 小时额度和每周额度也并不遵循同一套限制规则。这些都可能导致用户对额度的使用体验忽高忽低。 而 SemiAnalysis 最近披露的测试，则揭示了除上述因素之外的可能性： 即便购买相同套餐，不同账号的额度设置也可能存在差异。 Claude 的使用额度，究竟是怎么测出来的？ 用户感受到额度缩水，但 Claude 的使用上限究竟是多少，又是如何变化的？ 从进度条反推出真实的
Related stories
Meta, Microsoft reportedly cut employee use of Anthropic's Claude
According to The Information, Meta and Microsoft are working to reduce their employees' reliance on Anthropic's Claude, with Meta's Claude Code active users dropping to about 30,000 from 60,000 earlier this yearcars. Microsoft has cut internal Claude spending by over 33%, reducing per-employee monthly token budgets from $100,000 to $10,000 and routing requests to cheaper models via Copilot. Meta is shifting developers to its own tools like Muse Code and MetaCode.
3 reports · 3 sources375 points · 383 comments
SemiAnalysis: Anthropic subscriptions offer ~5x API-equivalent value vs OpenAI for agentic workloads
A SemiAnalysis analysis comparing AI subscription plans from Anthropic, OpenAI, Meta, SpaceXAI, MiniMax, Moonshot, Z.ai, Cursor, and Cognition found that Anthropic's subscriptions offer roughly five times more API-equivalent value per month than OpenAI's for agentic workloads, specifically comparing Claude Opus 5.5 against GPT-6.1 Sol. The analysis involved limit testing each subscription plan.
1 reports · 1 sources
Anthropic launches Claude Haiku 5.5 with effort controls, 90% price cut
Anthropic released Claude Haiku 5.5, its cheapest and fastest small model, priced at $0.10 per million input tokens and $0.50 per million output tokens for requests up to 100K tokens, a 90% cut from Haiku 4.5. It is the first Haiku model with adjustable reasoning effort, and benchmarks show it outperforming GPT-6 Luna on several tests, including OSWorld 2.1 and GDPval-AA v2.1.
6 reports · 5 sources1045 points · 487 comments
Anthropic: Claude flags 493 security issues across 25 open-source projects
Anthropic reported that Claude identified 493 security issues across 25 open-source projects, covering over 5 million lines of C, C++, Java, PHP, and Python code, including one vulnerability that went unnoticed for 21 years. The lab also announced a partnership with Theorem to use Claude for formally proving security properties at the assembly level across more than 100 real-world open-source libraries, extending formal verification to larger codebases.
News updates every hour. Subscribe in your RSS reader to get new stories automatically.
Subscribe