Claude Max promises "20x more". Two subscribers take Anthropic to court: 20 times WHAT? And the judge refuses to throw out the case

In short

• Alex G. Tse, a United States Magistrate Judge, denies in part Anthropic's motion, which sought to end the whole case. Surviving: the consumer-law claim for Solomon Smith, negligent misrepresentation and breach of contract. Dismissed for now, with leave to amend: the same consumer-law claim for Karl Kahn, two other California statutes and the good-faith claim.
• For the judge, a reasonable subscriber who pays monthly could plausibly read "5x" as per month. The "per session" clarification sat two hyperlinks away from the sales page. Anthropic answers that it was all written down, and that it never promised a fixed total amount of usage.
• The "3.5x" and "6x" figures (3.5 and 6 times Pro) are the plaintiffs' calculation. They start from hours estimated by Anthropic in emails of July 2025, quoted in the initial complaint: 140 to 280 hours of Sonnet 4 use per week for Max 5x, 240 to 480 for Max 20x, against 40 to 80 for Pro. More than the 168 hours in a week? Yes: the initial complaint attributes it to Claude Code sessions running in parallel. Nothing is proven. Next deadline: 02/12/2026, the date set for any amended complaint.

The label, the asterisk and two links

Four plans for Claude: free, Pro, Max 5x and Max 20x. Pro costs $17 to $20 (US dollars) per month, Max 5x $100, Max 20x $200. The three paid plans include Claude Code, Anthropic's coding tool, which the free tier lacks. Over a year, the bill comes to $1,200 for Max 5x and $2,400 for Max 20x, against $204 to $240 for Pro.

What separates the paid plans is usage. On claude.com, Max 5x promised "5x more usage than Pro", Max 20x "20x more usage than Pro". After "more usage", an asterisk pointed to one line: "Usage limits apply. Prices and plans are subject to change at Anthropic's discretion."

I followed the thread, as the order describes it. "Usage limits apply" is a link. It leads to a best-practices page. That page points, among others, to a "Max plan" link. At the end, a help page: "Max 5x provides 5 times more usage per session than the Pro plan", and Max 20x 20 times more. And Max also has weekly limits. According to the complaint, you can subscribe without ever opening that page: it is "outside the sales funnel". (When I write "the complaint" without more, I mean the consolidated complaint the judge examined. It is not public: I know it only through the quotes in the order and in Anthropic's motion.)

A session is a five-hour window. According to the help pages Anthropic cites in its motion, the per-session limit resets every five hours, the weekly limit every seven days.

And what unit is "usage" measured in? None of the court papers gives a precise one. Anthropic's help page describes a "conversation budget": how many messages you can send, or how long you can work with Claude Code, before the reset. The site shows the share of the limit already used, as a percentage. According to the initial complaint, it does not translate that into tokens, hours or queries.

The menu has changed, by the way. According to Anthropic's help page read on 01/11/2026, "The Max 20x plan is now called the Ultra plan" (our bill on the $200 Claude Ultra plan). The order does not mention it: it examines the advertising as the complaint describes it.

Five hours versus seven days

Here is the mechanism the judge describes, based on the complaint. A Max subscriber does get 5 or 20 times more usage per five-hour session. But, like everyone, they have a weekly limit. It is higher than Pro's, but not 5 or 20 times higher, according to the plaintiffs' calculation.

The result, the judge writes: a Max subscriber may hit the weekly limit sooner than a Pro subscriber, and a Max 20x subscriber sooner than a Max 5x one. According to the complaint, Karl Kahn found that a single five-hour window, used to the full on Max 20x, could burn 15% of his weekly allotment.

The plaintiffs' hours, redone by hand

Where do 3.5 and 6 come from? From estimates Anthropic sent by email on 28/07/2025, according to the initial complaint. One per plan. Pro: most users can expect 40 to 80 hours of Sonnet 4, an Anthropic model, per week. Max 5x: 140 to 280 hours. Max 20x: 240 to 480 hours. Anthropic's motion also describes the email to Pro subscribers and its 40 to 80 hours.

I redid the divisions. 140 ÷ 40 = 3.5 and 280 ÷ 80 = 3.5. 240 ÷ 40 = 6 and 480 ÷ 80 = 6. If "5x" applied per week, it would take 200 to 400 hours; for "20x", 800 to 1,600. The initial complaint goes further: Max 20x costs twice as much as Max 5x ($200 versus $100) for 1.7 times its hours (240 to 480 versus 140 to 280). That 1.7 is rounded: 240 ÷ 140 and 480 ÷ 280 both come to about 1.71. These are their calculations. The judge cites the 3.5 and the 6, taken from the consolidated complaint, as allegations taken as true at this stage, not as established facts.

Anthropic has its own bill. If Max 20x promised 20 times Pro's hours per week, that would be 800 to 1,600 hours. A week has 168 hours. No reasonable consumer, Anthropic argues, reads a promise of "something that is impossible".

A detail both sides know: the 240 to 480 hours of Max 20x already exceed 168. The initial complaint explains it, citing Anthropic's own email, by several Claude Code instances running in parallel. Anthropic answers that this proves its point: per-session usage depends on model choice, conversation length, complexity and settings. It adds that the emails went only to existing subscribers, so they "could not have induced any subscription decision".

Anthropic's position

Its motion of 22/09/2026 comes down to one idea: the plaintiffs ignore what Anthropic actually tells subscribers. "5x" and "20x" are counted per session, and it says so. The word "also" on the help page signals separate weekly limits. Clicking the links in the purchase flow is "the digital equivalent of flipping a product over to read the back label".

Second argument: Anthropic never promised a fixed total amount of usage. Its terms let it "increase or decrease capacity limits". And the plaintiffs do not allege they received less than 5 or 20 times the per-session usage of a Pro subscriber.

On the "50%" savings shown for Max 20x, Anthropic runs the numbers: 20 times Pro's per-session usage would take four Max 5x plans, $400 a month, versus $200 for one Max 20x. The initial complaint counts by the hour instead and finds an hour of Sonnet 4 dearer on Max 20x ($0.09 to $0.19) than on Max 5x ($0.08 to $0.16). Two calculations, two results. The judge does not evaluate this one separately. Anthropic asked for every claim to be dismissed, without leave to amend.

Why the judge does not close the file

First, the rules of the game. At this stage, the judge takes the facts alleged in the complaint as true. He does not weigh evidence yet. He only asks whether, on paper, the complaint holds up. And courts "rare[ly]" dismiss on this test at this stage: whether an ad deceives is usually a question of fact, to be settled later.

Then, two packages, from two federal appeals court decisions the judge cites. A shampoo labelled "Nature Fusion": the words mean nothing precise, so the consumer must read the back, and a judge can take the back label into account right away. Wipes labelled "plant-based": the words have a meaning, so the back label is not enough to dismiss the claim.

For the judge, claude.com's sales page looks like the wipes. You are on the page. You see $100 a month and "5x more usage than Pro". You pay monthly. You can reasonably read "per month", in his view.

The "per session" clarification, two hyperlinks from the sales page, "may persuade the fact-finder" that the advertising is not misleading, the judge writes. But it is not a basis for dismissal at this stage. As for the asterisk, its line "says nothing about usage limits per session".

Claim by claim

Source: order of 06/11/2026, pp. 18-19.

California consumer law (CLRA). It protects only "consumers", not commercial use. Solomon Smith, a computer-science student, uses Claude for coursework and projects: "Educational use is plainly not commercial use", the court agrees, so he is plausibly a "consumer". Karl Kahn, on the other hand, does not detail what he uses Claude for. The judge also rejects Anthropic's argument that a subscription is neither a good nor a service: Claude Code, which offers ongoing coding help, is plausibly a "service".

Unfair competition (UCL) and false advertising (FAL). These two statutes only provide for an injunction or restitution. The plaintiffs are still Max subscribers: no standing for an injunction for now. And they do not explain why damages would not be enough: no restitution either.

Negligence and contract. Negligent misrepresentation follows the same test as the CLRA, met at this stage. For the contract, the terms of service point to "the order process" for what the subscription includes. That process plausibly included the pages showing $100 a month and "5x more usage than Pro". The "per session" clarification, outside the funnel, plausibly was not part of it. And Anthropic's reservation over capacity limits does not help it at this stage: the plaintiffs do not claim a fixed amount, they claim the multiplier.

Good faith. This claim targets changes made "during billing cycles". The terms allow them. The contract is not illusory: a Max subscriber still gets more hours per week than a Pro subscriber, and Claude Code. And according to the complaint itself, Anthropic tightened limits because demand strained its computing capacity. For the judge, as the complaint describes it, that was not "a bait-and-switch".

What we don't know yet

• Whether the plaintiffs amend their complaint. The judge gives them until 02/12/2026. Without a new complaint, Anthropic must answer by 16/12/2026.
• Whether the class will be certified. According to Anthropic's motion, which summarises the consolidated complaint, the plaintiffs sue on behalf of all natural persons in the United States who bought or upgraded to Max 5x or Max 20x since 09/04/2025. The order speaks of a "putative" class, proposed but not certified; it does not rule on that point.
• Who is right on the merits. Was a reasonable subscriber misled? Do the plaintiffs' hours hold up? None of that has been decided.
• What the consolidated complaint of 08/09/2026 says, the one the judge examined. It is not public in the archive we checked: I only know it through quotes in the order and in Anthropic's motion.
• A public reaction from Anthropic to the order. None in the documents I read. Its position is the one in its motion.

Bottom line: the court did not say the advertising misleads. It said the complaint holds up well enough, on paper, to go on. Me, I read the bill: $100 or $200 a month, a "5x" or a "20x" in the shop window of the sales page, and the unit of measure tucked two links away. When people end up in federal court over what they're buying for up to $2,400 a year, to me the subscriber didn't misread: the label mumbled.

Key terms

• Motion to dismiss: a request asking the judge to end a case before any evidence is examined. The judge then takes the facts alleged in the complaint as true. Anthropic's motion relies in particular on Rule 12(b)(6) of the US Federal Rules of Civil Procedure (FRCP).
• Plausibly misleading: credible enough, on paper, for the case to go on. Not a ruling on the merits.
• Magistrate judge: the exact title of the judge who signed the order, Alex G. Tse (United States Magistrate Judge), a federal judge who here decides Anthropic's motion.
• Putative class action: a case brought on behalf of a group of customers the court has not yet certified.
• Consolidated complaint: a complaint that merges several cases on the same subject into one.
• CLRA (Consumers Legal Remedies Act): California's consumer remedies law.
• UCL (Unfair Competition Law): California's unfair competition law.
• FAL (False Advertising Law): California's false advertising law.
• Negligent misrepresentation: false information given without enough care, which a customer relied on.
• Implied covenant of good faith and fair dealing: a rule of California contract law that bars depriving the other party of what it bought, even where the contract does not say so.
• Injunction: a court order to do or stop something, for example change an ad.
• Restitution: giving back money obtained unfairly.
• Session: a five-hour window of Claude use, with its own limit.
• Claude Code: Anthropic's tool that writes code from the user's requests.
• Sonnet 4: an Anthropic AI model.
• AI: artificial intelligence.
• PBC (public benefit corporation): a US mission-driven company status.
• Et al.: Latin for "and others".
• UTC: Coordinated Universal Time.
• PDF: a document format.
• Python/PIL: a programming language and its imaging library.
• ,…: pointers to the source passages (S) and to our own readings or calculations (R), at the end of the article.

Source sheet

• Court: US District Court for the Northern District of California, Kahn et al. v. Anthropic PBC (court case identifier: 3:26-cv-05763-AGT).
• Order: document 50, 06/11/2026, signed by Alex G. Tse, United States Magistrate Judge, 19 pages.
• Anthropic: motion to dismiss, document 39, 22/09/2026, 34 pages.
• Plaintiff: Karl Kahn's initial complaint, document 1, 14/06/2026, 35 pages.
• Not read: the consolidated complaint (document 37, 08/09/2026), missing from the public archive.
• Link: Kahn v. Anthropic case docket on CourtListener

Source passages

Documents downloaded on 09/11/2026 around 13:23 UTC from CourtListener's public RECAP archive, read in full (dated copies in the article folder). Pages = PDF pages. Amounts in US dollars.

Show the 5 sources

Ref.	Quote (original language)	URL
S1↑1↑2↑3↑4↑5↑6+28	p. 1: "Karl Kahn and Solomon Smith allege that Anthropic PBC misleadingly markets its Claude Max subscriptions as providing “5x or 20x more usage” than Pro plans." ; "On behalf of a putative class of Claude Max subscribers" ; "The advertising in question is plausibly misleading, so the case will move forward. The Court will grant in part and deny in part the motion and give Plaintiffs leave to amend the dismissed claims." — p. 2: "Courts “rare[ly]” dismiss CLRA claims at the pleading stage for failure to satisfy the reasonable consumer test; deceptiveness is usually a question of fact." — p. 3: "[a]ll allegations of material fact in the complaint are taken as true" ; "four tiers of service: free, Pro, Max 5x, and Max 20x" ; "The Pro plan costs $17 to $20 per month, the Max 5x plan costs $100 per month, and the Max 20x plan costs $200 per month." ; "Because payments are due monthly, reasonable consumers would plausibly understand the phrases “5x more usage than Pro” and “20x more usage than Pro” to mean [...] 5x or 20x more usage than Pro per month. But they don’t." ; "get 5x or 20x more usage per five-hour session, but like other Claude users they are subject to weekly usage limits" ; "Max 5x users get 3.5x more usage than Pro users per week, while Max 20x users get 6x more usage than Pro users per week" ; "a Max subscriber may hit his weekly usage limit sooner than a Pro subscriber will" — p. 4: "could burn a whopping 15% of the weekly usage allotment" ; "“Usage limits apply. Prices and plans are subject to change at Anthropic’s discretion.”" ; "Usage limit best practices" ; "“Max 5x provides 5 times more usage" — p. 5: "per session than the Pro plan,”" ; "located two hyperlinks away from the primary sales pages" ; "“outside the sales funnel,”" ; "McGinity" — p. 6: "“plant-based”" ; "Claude.com’s sales pages are more like the front label in Whiteside than in McGinity." — p. 7: "“may persuade the fact-finder that [Anthropic’s advertising] is not misleading, but it is not a basis for dismissal under FRCP 12(b)(6).”" ; "The disclaimer says nothing about usage limits per session." — p. 8: "the Court won’t separately evaluate whether other Anthropic representations (e.g., a “50% savings” representation [...]) also pass the reasonable consumer test" ; "is plausibly a “service” under the CLRA" — p. 9: "it offers users ongoing coding help" — p. 10: "Smith is a computer-science college student" ; "“Educational use is plainly not commercial use,”" ; "As for Plaintiff Kahn, he doesn’t explain in any detail what he uses Claude to do." — p. 11: "denies Anthropic’s motion to dismiss Plaintiff Smith’s CLRA claim" ; "grants Anthropic’s motion to dismiss Plaintiff Khan’s CLRA claim, with leave to amend" — p. 12: "Plaintiffs remain Max subscribers." ; "Plaintiffs currently lack standing to pursue UCL and FAL claims for injunctive relief." — p. 13: "They allege only that they “have no adequate remedy at law,” without explaining why." ; "the Court dismisses the UCL and FAL claims, with leave to amend" — p. 14: "Plaintiffs have stated a plausible negligent misrepresentation claim." — p. 15: "“will be described in the order process.”" ; "Anthropic plausibly breached the parties’ agreement" — p. 16: "Plausibly, then, the per session disclaimer wasn’t incorporated into the Terms" ; "Plaintiffs, however, are not alleging that Anthropic promised to provide them with “a fixed amount of total usage;”" ; "Plaintiffs have stated a plausible breach of contract claim." — p. 17: "changes that Anthropic made to Max subscriptions “during billing cycles,”" ; "Anthropic exercised its contractual rights" — p. 18: "can still use Claude for more hours per week than Pro and free users can" ; "“strained Anthropic’s output capacity capabilities.”" ; "Anthropic didn’t tighten usage limits as a bait-and-switch." ; "The Court dismisses Plaintiffs’ implied covenant claim, with leave to amend." — p. 19: "Their amended complaint is due on December 2, 2026. If Plaintiffs don’t amend their complaint, Anthropic shall answer the operative complaint by December 16, 2026." ; "Dated: November 6, 2026" ; "Alex G. Tse United States Magistrate Judge"	https://www.courtlistener.com/docket/73482575/kahn-v-anthropic-pbc/
S2↑1↑2↑3↑4↑5↑6+11	p. 1: "(Consolidated with 4:26-cv-07789-AGT)" ; "Consolidated Complaint: Sept. 8, 2026" — p. 2: "Rules 12(b)(1) and 12(b)(6) of the Federal Rules of Civil Procedure" — p. 9: "Anthropic makes clear that its Max 5x and Max 20x plans provide 5 or 20 times more usage “per session” than the Pro plan" ; "Anthropic never promised a fixed total amount of usage or capacity." ; "Plaintiffs never allege (nor could they) that they actually received less than 5 or 20 times the amount of per-session usage that a Pro user received." — p. 10: "the Court should dismiss the Complaint without leave to amend" — p. 12: "the per-session usage limits, as opposed to the weekly usage limits, reset every five hours" ; "which reset every seven days" — p. 13: "“‘conversation budget’ that determines how many messages you can send to Claude, or how long you can work with Claude Code, before needing to wait for your limit to reset.”" ; "monitor the percentage of their limits that they have consumed" ; "Plaintiffs filed this putative class action on behalf of all natural persons in the United States who purchased or upgraded to a Max 5x or Max 20x plan from April 9, 2025 to the present" ; "in its July 28, 2025 emails" — p. 15: "Max plans also have two weekly usage limits" ; "The word “also” signals that the weekly usage limits are an additional, independent limit" — p. 16: "Anthropic communicated hours-per-week usage estimates to existing subscribers" ; "This July 2025 email explained to Pro subscribers that they could expect approximately 40–80 hours per week of Sonnet 4 usage" ; "the Max 20x plan would guarantee 20 times that amount—800 to 1,600 hours per week of usage—which exceeds the 168 total hours that exist in a week by a factor of roughly five to ten" — p. 17: "something that is impossible" ; "It was sent only to then-current subscribers and thus could not have induced any subscription decision." ; "But that defeats rather than rescues their theory." ; "multiple factors, including model selection, conversation length, computational complexity, and configuration settings" ; "the digital equivalent of flipping a product over" — p. 18: "to read the back label" — p. 19: "“may sometimes add or remove features, increase or decrease capacity limits”" — p. 20: "Max 20x provides 20 times the Pro plan’s per-session usage at $200 per month." ; "To obtain equivalent per-session usage, a subscriber would need to purchase four Max 5x subscriptions at $100 each, totaling $400 per month." — p. 33: "the Court should dismiss all of Plaintiffs’ claims with prejudice"	https://www.courtlistener.com/docket/73482575/kahn-v-anthropic-pbc/
S3↑1↑2↑3↑4↑5↑6+4	p. 14: "it does not quantify the usage by reference to tokens used, hours spent, or queries made" — pp. 14-15: "in emails to subscribers on or around July 28, 2025" ; "“Most Pro users can expect 40-80 hours of Sonnet 4 within their weekly rate limits." ; "“Most Max 5x users can expect 140-280 hours of Sonnet 4" ; "“Most Max 20x users can expect 240-480 hours of Sonnet 4" — p. 16: table 1: "Max 5x Sonnet 4 140-280 200-400 3.5x more usage than Pro" ; "Max 20x Sonnet 4 240-480 800-1600 6x more usage than Pro; 1.7x more usage than Max 5x" ; footnote 1: "There are only 168 hours in a calendar week. As Defendant’s email points out, subscribers can initiate “multiple Claude Code instances in parallel,” presumably accounting for the number of hours in the weekly limit exceeding 168." — p. 17: table 2: "Max 5x Sonnet 4 140-280 hours $100 $0.08-$0.16" ; "Max 20x Sonnet 4 240-480 hours $200 $0.09-$0.19"	https://www.courtlistener.com/docket/73482575/kahn-v-anthropic-pbc/
R1↑1↑2	Our reading of 01/11/2026, 17:51 UTC (article no. 95): "The Max 20x plan is now called the Ultra plan. Existing Max 20x subscribers move to Ultra automatically, with no price change."	https://support.claude.com/en/articles/17728952
R2↑1↑2↑3	Our calculations: $100 × 12 = $1,200; $200 × 12 = $2,400; $17 × 12 = $204 and $20 × 12 = $240; 140 ÷ 40 = 280 ÷ 80 = 3.5; 240 ÷ 40 = 480 ÷ 80 = 6; 240 ÷ 140 ≈ 480 ÷ 280 ≈ 1.71 ("1.7x" in the initial complaint).	—