Anthropic is facing a federal lawsuit from a paying customer who says the company misled him about how much he could actually use its premium Claude subscriptions. Karl Kahn, a Washington DC resident, filed the suit this week accusing Anthropic of overselling the usage limits on its Max plans. According to Engadget, the case was first reported by The Wall Street Journal.
The lawsuit claims that the real usage caps on Anthropic’s Max 5x and Max 20x tiers are difficult to pin down and appear to fall well short of what the company advertises. Kahn upgraded to the Max 20x plan in April after starting to use Claude Code, Anthropic’s AI coding tool. He says he burned through 15 percent of his weekly allowance in a single five-hour session. The lawsuit is seeking class-action status on behalf of other US consumers who bought a Max plan since Anthropic started offering them last year.
Anthropic declined to comment on the case.
To understand why this matters, it helps to know how Anthropic structures its paid plans. The company offers three individual subscription tiers:
- Claude Pro at $17 per month, which promises at least five times the usage of the free tier during peak hours
- Max 5x at $100 per month, offering up to five times the usage of Pro
- Max 20x at $200 per month, promising up to 20 times the Pro allowance
Even at those prices, the company’s own website acknowledges that message counts vary based on length, file attachments, conversation history, and which model or feature you use. That kind of vague language is exactly what the lawsuit takes issue with. “The actual usage provided by the Max 5x and Max 20x plans is far below the advertised amount of usage,” it alleges.
Complaints about rate limits are nothing new in Anthropic’s user community. Reddit threads on the topic come up regularly, with one recent post describing how a single Claude Code prompt ate through an entire five-hour usage window. Anthropic introduced weekly rate limits on Claude Code back in July 2025 after some users were running the coding agent around the clock in the background.
The core issue here is a technical one. Every AI model runs on tokens, the units it uses to process text. When you type a prompt, the model breaks your words, punctuation, and characters into numbers that connect to patterns it learned during training. Generating a response has both input and output costs, and those costs can spike fast depending on how complex your request is. A simple question is cheap. A long coding session is not.
This is where the lawsuit points to something much bigger than one unhappy subscriber. Traditional software subscriptions sell access to a fixed product. You pay for Spotify and you get Spotify. But AI inference doesn’t work that way. The compute cost of running a model scales with how you use it, and that cost can swing dramatically from one session to the next. Flat-rate pricing for something this variable was always going to create friction.
Right now, venture capital is helping companies like Anthropic absorb the gap between what users pay and what it actually costs to run these models. But that situation won’t last forever. Both Anthropic and OpenAI are expected to go public eventually, and when they do, the pressure to close that gap will increase sharply. The question of how to honestly communicate usage limits to paying customers isn’t just a legal problem. It’s a business model problem the entire industry will have to answer.
