Anthropic's Max Plan Faces Lawsuit Over Undisclosed Weekly Usage Caps

Anthropic advertises its top-tier Claude Max subscription as offering 20 times the usage of its $20-per-month Pro plan for $200 monthly. Yet customers paying for Max—whether the $100 "5x" or $200 "20x" tier—still encounter a separate weekly ceiling that wasn't prominently disclosed. An expanded class-action lawsuit filed this week argues the company failed to make these restrictions sufficiently clear before purchase.

The case highlights a fundamental tension in AI pricing: vendors must fit inherently unpredictable computational costs into fixed monthly fees, yet the actual value customers receive can swing wildly depending on their workload. When usage guarantees come with hidden strings, clarity becomes a competitive and legal issue.

The Fine Print Behind "20x More"#

Anthropic's public documentation indicates that its usage multipliers apply only within five-hour sessions and reset at those intervals. However, the company also enforces a broader weekly cap across all models—and reserves the right to add further capacity restrictions without notice. For Claude Code users running interactive sessions, the complexity deepens: those sessions count against the same plan limits, meaning a developer's actual runway depends on how many and how long they run.

According to the lawsuit, which first surfaced in The Verge, Anthropic introduced these weekly restrictions in late July 2025, months after Max launched in April. The company continued marketing the same 5x and 20x multipliers without adequately highlighting the new constraints. Anthropic's defense—that customers could find this information via hyperlinks during checkout, similar to reading a product label—hasn't swayed the plaintiffs, who say the marketing premise itself was misleading.

The Broader Pricing Problem in AI#

Claude Max is far from alone in struggling to translate compute into clear value. When tasks range from simple bug fixes to complex multi-hour debugging sessions, a usage multiplier doesn't meaningfully predict how long a developer can work. That ambiguity extends across the industry.

OpenAI faces the same challenge with its Codex product and other agentic tools, which draw from a shared allowance. The company attempts to address it by noting that usage varies by task, model, and execution context—and customers can purchase additional credits. OpenAI also varies pricing by model; Astra, for instance, costs 2.5x more per token than GPT-5.6, making the same budget stretch or compress significantly depending on which model a team chooses.

Across all these platforms, the core problem persists: saying a plan offers "substantially more" usage doesn't tell developers whether they can finish their sprint, debug a critical system, or run their agents overnight without hitting a paywall.