A Washington, D.C. resident has filed a federal lawsuit against Anthropic, accusing the company of misleading consumers about the usage limits on its premium Claude Max plans. The suit, which seeks class-action status, was lodged on Monday by Karl Kahn, who claims the actual caps on Anthropic’s highest-tier subscriptions fall short of what the company advertises.
Anthropic offers three paid tiers for individual users. The entry-level Claude Pro plan costs $17 per month and, according to the company, provides at least five times the usage per session compared to the free service during peak hours. Anthropic notes on its website that message limits vary based on factors like message length, attached files, and the specific model or feature in use.
Related: Why purple USB ports are rare in America
Since April 2025, the company has also sold two higher-tier plans: Max 5x and Max 20x. Priced at $100 and $200 per month respectively, these plans promise up to five and 20 times the usage caps of the Pro package. The lawsuit alleges these promises don’t hold up in practice.
“The actual usage provided by the Max 5x and Max 20x plans is far below the advertised amount of usage,” the complaint states. Kahn upgraded to the Max 20x plan in April after he started using Claude Code, Anthropic’s coding assistant. He quickly hit his weekly usage limits. In one five-hour session, he burned through 15 percent of his weekly allowance, according to the filing.
Related: Best shots in cricket history, All cricket shots name with image
The lawsuit seeks class-action status on behalf of other U.S. consumers who bought a Max subscription since the plans launched last year.
Rate limits on Anthropic’s services have been a recurring gripe on forums like Reddit. One user recently complained about exhausting a five-hour limit after a single Claude Code prompt. Last July, the company introduced weekly rate limits on Claude Code usage, citing users who had been running the coding agent “continuously in the background, 24/7.”
Related: Engadget website error disrupts access
All large language models operate on a unit called tokens. When a user types a question into Claude, the system converts words, character groups, and punctuation into numbers that map to patterns the model learned during training. Generating a response requires both input and output processing, and those costs can vary significantly depending on how complex a prompt is.
The lawsuit shows a growing disconnect between what consumers expect from software subscriptions and the actual cost of running AI inference at scale. Traditional subscription models don’t translate neatly to the compute demands of large language models. Right now, venture capital is subsidizing those costs, and the problem is likely to get worse when Anthropic and OpenAI eventually go public.
