A class-action lawsuit filed Tuesday accuses Anthropic of failing to clearly disclose usage restrictions on its Claude Max subscription tiers, which cost $100 and $200 per month and promise five times and 20 times more usage than the company's $20 Pro plan. The expanded complaint alleges that developers purchasing the Max 20x tier can still hit a separate weekly usage ceiling that wasn't adequately explained at signup. According to The New Stack's report, the case highlights a broader challenge facing AI companies: packaging unpredictable compute demands into fixed monthly subscriptions that customers can easily understand.
Anthropic's documentation reveals that the 5x and 20x multipliers apply on a per-session basis, with limits resetting every five hours. But that five-hour window isn't the only constraint, since the company also enforces a weekly usage cap across all models and reserves the right to impose additional restrictions to manage capacity. The complaint states that Anthropic introduced these weekly limits in late July 2025, months after Claude Max launched in April, while continuing to advertise the plans using the 5x and 20x usage claims. Anthropic has countered that customers could find information about the limits through hyperlinks during the purchase process, comparing those disclosures to details on a product label that shoppers can review before buying.
The report quotes Sajid Afridi, CTO of Pakistan Red Team and an enterprise systems architect, who says "a single autonomous debugging loop can consume millions of tokens via context re-submission and tool calling," meaning a subscriber can burn through their entire weekly quota in just a couple of intense work sessions. Manish Jain, founder and principal analyst at Strategic Horizon, notes that if he sees a SaaS product advertised as offering five or twenty times more usage, his natural interpretation would be that he can accomplish roughly five to twenty times as much work. But with AI, capacity hinges on the model, context, and workload, making it hard to predict what those higher limits will actually deliver for any given developer.
The lawsuit points to a fundamental tension in how AI coding subscriptions are priced. Because coding tasks vary so dramatically, a usage multiplier doesn't tell developers much about the difference between a quick fix and a complex job that can drain an allowance much faster. OpenAI faces similar challenges with Codex, where usage depends on the task, the model, and where the work runs, and where model choice matters significantly since Astra costs 2.5 times more per token than GPT-5.6. The report notes that OpenAI has been testing outcome-based pricing with some enterprise customers, charging only when an agent completes a task rather than metering raw compute, though that approach introduces its own problems around defining success.
If the plaintiffs win, other AI providers may need to rethink how they describe their subscriptions, particularly around when usage resets and what restrictions apply. But even more detailed disclosures at signup don't answer what developers really want to know: how much work they can complete for the price they're paying. As coding agents take on larger jobs and operate autonomously for longer periods, the case could establish how much information providers must disclose when selling subscriptions whose actual value can swing wildly depending on the workload. The New Stack reached out to Anthropic for comment on the lawsuit and its Claude Max usage policies but had not received a response at the time of publication. Subscription models that rely on abstract multipliers rather than transparent capacity guarantees may struggle to satisfy both regulatory scrutiny and customer expectations as AI tools become mission-critical infrastructure. Providers who can't bridge the gap between marketing simplicity and operational reality risk losing trust exactly when enterprise adoption demands it most.