A Claude subscriber has sued Anthropic, alleging the company mislead consumers about the usage limits of its Max plans. Per The Wall Street Journal, Karl Kahn, a Washington DC-based resident, filed a federal lawsuit on Monday accusing Anthropic of overselling how much he could use Claude. The lawsuit claims that the actual usage caps of Anthropic’s Max 5x and Max 20x tiers are hard to determine and appear lower than the limits the company advertises on its website and elsewhere.

Anthropic offers three different paid tiers for individuals. The company’s most affordable plan, Claude Pro, starts at $17 per month and “offers at least five times the usage per session compared to our free service” during peak hours. “The number of messages you can send will vary based on message length, including the length of files you attach, the length of your current conversation, and the model or feature you use,” Anthropic writes on its website. Since April 2025, the company has also offered Max 5 and 20x plans, which promise up to five and 20 times higher usage caps than the Pro package. The two plans cost $100 and $200 per month, respectively.

“The actual usage provided by the Max 5x and Max 20x plans is far below the advertised amount of usage,” the lawsuit alleges. After he began using Claude Code, Khan upgraded to the Max 20x plan in April of this year, but quickly found himself hitting his weekly usage limits. One five-hour session saw him go through 15 percent of his weekly allowance. The lawsuit is seeking class action-status on behalf of other US consumers who purchased a Max subscription since Anthropic began offering them last year.

Anthropic declined to comment.

Anthropic’s rate limits are a frequent topic of discussion on Reddit, with one user recently complaining about how they blew past their five-hour limit after just a single Claude Code prompt. Last July, the company imposed weekly rate limits on Claude Code usage in response to users who had been running the coding agent “continuously in the background, 24/7.”

All large language models operate on a currency known as tokens. When you type a question into Claude’s prompt bar, it converts words, groups of characters and punctuation into numbers that map to different patterns and relationships Anthropic’s models learned in training. For a model to generate a response, there are both input and output costs, and those can vary greatly depending on the complexity of a prompt.

The lawsuit highlights a long-growing disconnect between consumer expectations and the actual cost of providing AI inference. Traditional software subscription models don’t map neatly, if at all, to how much compute LLMs require to work at scale. Right now, venture capitalists are offsetting those costs, and if anything, the problem is likely to become more acute when Anthropic and OpenAI go public.