A class action filed on September 8 in the Northern District of California argues that Anthropic sold Claude Max on usage multipliers the service never delivers. The plaintiffs put a number on the gap: the 200 dollar tier advertises twenty times the usage of the Pro plan and reportedly hands over six to eight. Anthropic has answered with a motion to dismiss.
Key Takeaways
- The suit targets both Claude Max tiers, at 100 and 200 dollars a month, covering every purchase made since April 2024.
- The dispute sits on the five-hour session window and the weekly cap, which both shave down the advertised multiplier.
- Anthropic wants the case thrown out, arguing the usage terms sat behind links during checkout.
Have an AI Sum Up This Article
ChatGPTWhat the complaint says Claude Max actually delivers
The named plaintiff is Karl Kahn, a developer who moved to the top tier for heavy coding work and filed an initial federal complaint back in June. The expanded September 8 version asks for class status covering everyone who bought either plan since April 2024.
Both tiers are described plainly by the vendor itself. On the help page that lays out how the Claude Max plan works, Anthropic writes that Max 5x provides five times more usage per session than Pro, and that Max 20x provides twenty times more. Those two sentences sit at the center of the case.
Against them the filing sets measured usage on Claude Max. The 200 dollar tier reportedly lands between six and eight times Pro. The 100 dollar tier reportedly lands at three and a half. In both cases the distance from the advertised figure is wider than any rounding a marketing team could defend.
One reading from the plaintiff captures the experience better than the ratios do. A single five-hour work session burned roughly 15 percent of his weekly quota, which leaves under seven sessions of that intensity in a full week. Claude Code weekly limits drop another 17 percent on September 14, two days after the filing landed.
Damages for Claude Max buyers are pegged above five million dollars, attorney fees excluded. That threshold matters procedurally, since it is the line that opens federal jurisdiction for this kind of action, and it signals that the plaintiffs are chasing a broad class rather than a private settlement.
Anthropic has not disputed the mechanics themselves. The company moved to dismiss on the ground that the relevant details were reachable through hyperlinks during the purchase flow. Its defense therefore rests on what the buyer could have known, not on what Claude Max hands over.
That framing shapes everything that follows. It leaves the six-to-eight figure and the method behind it untouched, and moves the fight onto disclosure ground instead. A judge will first have to decide whether a link buried in checkout can qualify a number printed across a pricing page.
The five-hour window that eats the multiplier
Nothing about the contested mechanism is hidden, and that is exactly what makes the case worth watching. On Claude Max a usage multiplier does not apply to a month of subscription. It applies to a session, and that session resets every five hours.
Sitting above that window is a second ceiling, this one weekly, resetting at a fixed hour assigned to each account. Someone who packs their workload into two days does not get the same volume as someone who spreads it out, though both pay the same price for the same Claude Max tier.
The help page goes further still. Anthropic reserves the right to limit usage in other ways, weekly and monthly caps included, plus restrictions by model and by feature, at its own discretion. That clause turns the advertised multiplier into a revisable order of magnitude.
Recent history shows the clause gets used. Anthropic cut Fable 5 limits on Max and Team Premium on July 20 without touching the subscription price. The same lever moved the other way when Claude Code limits doubled for the SpaceX engineering teams.
For a subscriber the practical consequence fits in one line. You buy a usage right whose value depends on how you work and on the vendor’s infrastructure calls that month. A developer running long sessions hits the ceiling well before someone asking three questions a day, and neither of them can read that difference off the pricing page before paying.
Nothing in the record establishes yet that the plaintiffs’ numbers hold. The six-to-eight figure comes from their own measurements rather than an adversarial audit, and discovery exists precisely to test it.
A technical problem is waiting for both sides anyway. Measuring real usage on Claude Max means fixing a unit of account, and a consumed token shifts with the model called, the context length and any long reasoning pass. Two users can drain the same allowance on workloads that have nothing in common.
More articles on Horizon
- DeepSeek V4.1-Flash Cuts Agent Memory Eightfold
- This ChatGPT Guide Covers Real Professional Use
- ChatGPT Pro Has Stopped Taking New Subscribers
Why this case reaches every AI subscription
The timing lands badly for Anthropic on price. Claude Fable 5.1 shipped in early September with a 25 percent cost cut, and the company has spent the summer defending an aggressive stance on coding workloads.
That stance leans on public figures. Claude Opus 5 was pitched as matching the top of the market at half the cost. A lawsuit about the honesty of a multiplier therefore hits the central argument of the whole catalog.
The question runs past one vendor. Selling compute as a flat subscription means converting tokens into a readable promise, and the multiplier is the simplest formula anyone has found for that job. If a court calls that formula misleading, the marketing grammar of the entire sector changes.
Rivals are watching from a position that is no safer than Anthropic’s. OpenAI segments its tiers the same way, with quotas that shift by model and by window, and the recent freeze on sign-ups to its priciest plan was a reminder that available capacity dictates the commercial promise.
Teams holding a budget have a call to make before the case resolves. Staying on a Claude Max subscription means accepting a moving ceiling, while shifting to the API means accepting a variable invoice. Neither option restores the predictability the flat plan was supposed to buy.
The cheapest exit for the industry would be tougher disclosure rather than a retreat on price. Publishing a token volume, a session length and a weekly cap on the same grid costs less than litigation, though it also makes cross-vendor comparison far more brutal.
The next signal to watch is the ruling on the motion to dismiss. If the judge lets the case proceed, discovery will force Anthropic to document what Claude Max actually delivers, tier by tier. No generative AI vendor has ever had to put those numbers in front of a court, and the filing would become a reference point for every subscription dispute that follows.
Follow the story on Horizon.
