Claude Max's 20x promise draws a class action

An expanded class-action lawsuit filed Tuesday alleges that Anthropic did not adequately disclose a separate weekly limit on its $200-a-month Claude Max 20x subscription, despite marketing the tier as offering 20 times more usage than the $20 Pro plan. The case centers on how customers encounter plan limits during purchase.

The complaint was first reported by The Verge. The New Stack examined the terms that sit alongside the five-hour session reset, including a weekly cap that covers all models and Claude Code use. The dispute concerns whether those limitations were disclosed clearly enough before developers subscribed.

At a glance

• Claude Code’s interactive sessions count against the shared Max allowance, so a weekly limit can interrupt coding even when the plan’s per-session reset has not yet restored capacity.
• Anthropic says Max 5x costs $100 per month and Max 20x costs $200, while their advertised multipliers apply to usage within sessions that reset every five hours.
• Anthropic may impose further capacity restrictions, and its prior dismissal motion said buyers could access information about the limits through purchase-page hyperlinks before deciding whether to subscribe.

AI companies are putting variable compute consumption behind simple monthly prices, an arrangement that appears increasingly difficult to express through a single multiplier. A brief repair and a complicated coding assignment can draw down capacity at different rates. The suit therefore appears to test whether a comparison with another plan conveys enough about restrictions that may arrive before a subscriber’s work is complete.

Anthropic’s Max limits reset every five hours but also include a weekly cap.

Anthropic’s documentation says the Max 5x and Max 20x multipliers apply per session, with usage limits resetting every five hours. It also says a weekly usage cap applies across all models and that further restrictions may be added to manage capacity. Claude Code’s interactive coding sessions count toward those same plan limits.

According to the complaint, the company introduced the weekly cap in late July 2025, months after Max launched in April, while continuing to market 5x and 20x claims. The plaintiffs say the constraint was not adequately disclosed at subscription. Anthropic’s motion to dismiss an earlier version said customers could reach limit information through hyperlinks during purchase.

Codex usage depends on the task, model, and where work runs.

Once included usage is exhausted, developers can wait for a reset or pay more to keep working. The Max multiplier compares the tier with Pro, but it does not set a fixed quantity of coding work for a given subscription. Task complexity can also affect how quickly a coding job consumes an allowance.

OpenAI’s Codex documentation says usage depends on the task, model and where work runs. Codex can draw from an allowance shared with other agentic products, and subscribers may buy more credits after exhausting it. OpenAI has also made Codex more autonomous, allowing work to continue while it awaits a developer response.

The model selected changes the rate of allowance consumption. Astra costs 2.5 times more per token than GPT-5.6, so the same allowance can last for markedly different amounts of work. The lawsuit does not accuse OpenAI or other providers of wrongdoing.

OpenAI is testing outcome-based pricing with some enterprise customers.

OpenAI has been testing outcome-based pricing with some enterprise customers, charging when an agent completes a task instead of metering raw compute. The model requires a definition of success, and failed agent runs become the provider’s expense rather than a customer charge tied directly to compute consumption.

The mechanics differ from Claude Max’s structure. In its earlier motion, Anthropic compared its linked disclosures to details on a product label that customers can find by turning over a package before deciding whether to buy it at purchase.

What the weekly-cap claim tests If the plaintiffs succeed, the case could establish a clearer standard for how AI providers describe subscription restrictions at signup, particularly resets and caps. That would not settle the workload question on its own: coding tasks, model choice and agent behavior can still produce divergent consumption. The complaint’s resolution may instead define the disclosure baseline for plans built around comparative usage claims.