Claude Max in court: judge finds "5x more usage" claim was misleading

Claude Max in court: a judge finds "5x more usage" was misleading

🔎 A "20x" that was actually worth 3.5x to 6x

On November 6, 2026, Magistrate Judge Alex Tse, of the U.S. District Court for the Northern District of California, issued a 19-page order in the case Kahn v. Anthropic (26-cv-05763-AGT). At the heart of the matter: Claude Max's marketing slogan, "5x or 20x more usage than Pro." In the judge's view, that promise is plausibly misleading — and the false advertising claims can therefore proceed to trial.

The problem is disarmingly simple. Consumers pay monthly. They therefore reasonably read "20x more usage" as a monthly promise. But the multiplier actually applied to the 5-hour session, and the weekly caps — introduced after the fact — didn't scale up in the same proportions. The plaintiffs' math: the $200/month Max plan delivered only 6x the Pro plan's weekly usage, and the $100/month plan a mere 3.5x.

Why does this matter now? Because this ruling arrives at a moment when all the major AI labs — Anthropic, OpenAI, Google — are selling subscriptions with opaque quotas, "usage limits" never publicly quantified, and marketing ratios that flatter the offering. If a federal judge holds that this kind of messaging can constitute false advertising, the entire industry will have to revisit its sales pages. And users, for their part, finally have a framework for re-reading the fine print.

Key takeaways

• Judge Alex Tse (N.D. Cal.) declined to dismiss the negligent misrepresentation and breach of contract claims against Anthropic, concerning Claude Max's "5x / 20x more usage than Pro" promises.
• The multiplier applied to the 5-hour session, not the month: plaintiffs calculate 3.5x ($100 plan) and 6x ($200 plan) in actual weekly usage compared to Pro.
• The per-session disclosure was "outside the purchase funnel": it was "two hyperlinks away from the main sales pages," according to the order made public on govinfo.gov.
• Four claims are dismissed (including the California unfair competition and false advertising laws), with leave to amend; three survive, including breach of contract.
• The weekly caps date from late July 2025, i.e., after Max's launch in April 2025 — a point that fuels the sense of having bought a product different from the one advertised.
• The class has not yet been certified, and Anthropic has not commented on the ruling.

Recommended Tools

Before diving into the legal details, a quick reminder of the plans involved. The prices below are those mentioned in the court arguments and reported by Courthouse News — always check claude.ai before subscribing, as they change over time.

And if you're torn between competing assistants, our Claude vs ChatGPT comparison details the differences in real-world usage beyond the marketing slogans.

What Judge Tse Actually Decided

Direct answer: three claims survive (negligent misrepresentation, breach of contract, and Smith's CLRA claim), four are dismissed with leave to amend.

The order, available as a PDF on govinfo.gov, does not say that Anthropic lied. It says something more subtle and more dangerous for the company: that a reasonable consumer could have been misled. That is the legal standard of "plausible misleadingness" — and it is enough for the case to move forward.

The judge's reasoning boils down to three points. First, consumers pay monthly, so reading "20x more usage" as a monthly promise is reasonable. Second, the multiplier actually applied to the 5-hour session — information buried in a page accessible "two hyperlinks away from the main sales pages," described as "fine print" outside the purchase funnel according to Glitchwire. Finally, the asterisks referring to "Usage limits apply" are not enough to correct the impression created by the prominently displayed figure.

On breach of contract, the argument is clever: the terms of service refer to "the order process". Now, this process plausibly includes the sales pages. The page explaining the per-session limit, however, sits outside this process — it is therefore not part of the contractual agreement, as reported by Sovereign Magazine. Anthropic therefore cannot hide behind terms that the subscriber never saw at the time of payment.

Worth noting: the judge did, however, dismiss the claim that limits were tightened mid-billing-cycle. And Kahn's CLRA claim was dismissed with leave to amend, while Smith's survives — the court finding that Claude Code can qualify as a "service" within the meaning of California's consumer protection law.

Why the "5x / 20x" math never held up

Direct answer: because the multiplier applied to the session, but the weekly limits — the ones that actually matter — never followed the same proportion.

Let's take the plaintiffs' numbers, detailed in the court order. For the same model:

The gap is glaring on the premium plan: promised twenty times, delivered six times. The customer testimony cited by Courthouse News illustrates the lived experience behind these numbers: "50% of my weekly quota consumed by day 3 on the 20x plan." And Kahn, the lead plaintiff, burned 15% of his weekly quota in a single 5-hour session on the 20x plan.

Anthropic's internal logic is understandable: a 5-hour session technically cannot consume 20 times a month of Pro usage. The "20x" described a rate, not a volume. But that's precisely where marketing and reality parted ways. A customer who works in short, frequent sessions — the typical case of a developer running Claude Code all day — can never come close to the promised multiplier, because their limiting factor becomes the weekly cap, not the session.

My opinion, and I'm not alone in this: it's not fraud, it's deliberate ambiguity. Choosing the metric that produces the biggest number — the session rather than the month — is a copywriting technique as old as selling itself. Judge Tse has just reminded everyone that it has legal limits.

How customers really read LLM subscription plans

Direct answer: they read the big number, pay by the month, and discover the limits by hitting them.

That's the most interesting point of the case for anyone following the industry. AI labs' offerings all share the same marketing DNA: a catchy multiplier ("5x", "20x", "unlimited"), an asterisk, and a documentation page nobody reads before buying. The Kahn case shows what happens when a court looks at this architecture through the eyes of an ordinary consumer.

Three takeaways emerge from the ruling:

1. The payment point sets the unit of reading. You pay by the month? The judge holds that you can reasonably read the numbers by the month. It's a precedent that compels aligning the unit of the promise with the unit of billing.

2. Documentation doesn't save the marketing. The exact information existed — it was two clicks away. The judge considers two clicks too many. Disclosure must be inside the purchase funnel, not next to it.

3. Caps introduced after the sale create a risk of their own. Weekly caps arrived in late July 2025, after Max launched in April 2025. Selling a plan, then changing its effective limits a few months later, feeds exactly the kind of complaint seen here — even though the judge rejected that specific claim at this stage.

This isn't a problem unique to Anthropic. Opaque quotas are the industry norm, and usage limits run through every comparison — it's a criterion we cover in detail in Google Gemini vs ChatGPT vs Claude: which one for which use case?. The difference is that Anthropic is the first to be caught up with by a federal court on this specific ground.

Context: caps born of compute scarcity

Direct answer: the weekly caps are a direct consequence of the pressure on computing capacity, and they were introduced without realigning the marketing.

Let's go back to the timeline, because it sheds light on the whole affair. April 2025: launch of Claude Max, with its promises of 5x and 20x. Late July 2025: introduction of the weekly caps, reported by Courthouse News. In between, agentic usage exploded — Claude Code, automated workflows, long sessions — and the operating cost of the models followed.

This context is documented more extensively elsewhere: a METR study suggests that AI is accelerating the development of AI itself, and that Anthropic is progressing 19x faster on certain benchmarks thanks to its own tools. The more models are used to produce models, the more critical compute demand becomes — and the tighter the quotas get to protect the margin.

The paradox is cruel: the most loyal customers — those paying $200/month to push the tool to its limits — are precisely the ones who hit the caps the fastest. They are also, now, the plaintiffs. Judge Tse dismissed the grievance about limits being tightened mid-billing-cycle, but the mere existence of these caps, added after the sale, remains part of the case file for the initial deception.

Regulatory pressure is also building on several fronts for the California labs. XAI, for instance, faces a complaint under the Clean Air Act over its Colossus 2 power plant that the judge refused to dismiss. Between the environmental impact of data centers and business practices, California courts are becoming a central player in AI governance — in the absence of federal legislation worthy of the name.

Impact on Competitors' Offerings

Direct answer: if Tse's reasoning holds, every multiplier-based plan on the market becomes a potential target — and LLM marketing will have to put honest numbers behind its claims.

Look at the landscape through the eyes of a plaintiff's lawyer. OpenAI sells plans with "usage limits" that are never publicly quantified. Google does the same with Gemini. "Unlimited" or "priority access" style promises are everywhere. No player publishes the number of requests, tokens, or included compute hours — the opacity is structural.

What changes with the November 6 order is that a federal judge has validated an interpretive framework: the unit of the promise must match the unit of the billing, and disclosures must be part of the purchase process. Applied mechanically, this test would eliminate a good chunk of the industry's current marketing.

Three scenarios are taking shape:

• Cautious status quo: the labs add explicit disclaimers ("actual weekly usage varies with load") in the purchase funnel. This is the most likely short-term scenario, and the least costly.
• Forced transparency: publication of concrete numerical ranges (messages per week, included session hours). This would be a genuine consumer win, but it would expose the real differences between plans — and between competitors.
• Windfall effect: a less exposed player (DeepSeek, Kimi, the open-weight offerings) uses transparency as a selling point against the American giants.

For the end user, the immediate lesson is to compare plans on actual usage, not on multipliers. Our guide to the best LLMs for coding ranks models by real-world capability — Gemini 3.1 Pro leading overall, GPT-5.5 dominant in agentic tasks with a score of 98.2 — but choosing a subscription depends as much on quotas as on the model. And for a direct comparison of the two most widely used ecosystems, Claude vs ChatGPT remains our reference.

What Comes Next: Class Certification and Amendments

Direct answer: the class is not certified, Anthropic has not commented, and the plaintiffs can amend the dismissed claims.

Three steps remain before the court. The first is class certification: for the case to reach industrial scale, the judge must find that all Max subscribers since April 2025 share common questions of fact and law. Given the number of Max subscribers and the uniformity of the marketing, this step seems within reach — but nothing is guaranteed.

The second is discovery. This is where Anthropic faces the greatest risk. The plaintiffs will be able to demand internal consumption data, pre-launch marketing documents, and above all the internal analyses that preceded the introduction of the July 2025 caps. If an internal email shows that the marketing team knew the "20x" did not match actual weekly usage, the defense collapses.

The third is the right to amend: four claims were dismissed with leave to replead, giving the plaintiffs a second chance on California's unfair competition and false advertising laws — often the most financially punitive ones.

The most likely outcome remains a settlement. Anthropic has no interest in a public trial over its marketing while it is chasing enterprise customers, and the plaintiffs have already secured the essentials: the principle that a federal judge finds their reading plausible. But even a settlement will have a disciplinary effect on the entire industry, as reported by epoch0.tokyo's monitoring of AI legal news.

❌ Common Mistakes

Mistake 1: Believing the multiplier without checking the unit

"20x more usage" means nothing without knowing what the multiplier applies to: session, day, week, month? In the Claude Max case, the gap between the promise (20x) and the weekly reality (6x) came precisely from this ambiguity. The solution: look for the limits documentation page before paying, and convert the figure yourself into weekly usage.

Mistake 2: Assuming the limits are fixed over time

Claude's weekly caps arrived three months after the plan launched. AI subscription quotas evolve with server load, with no obligation to give notice. The solution: assume that today's plan may be more restrictive in six months, and read the terms modification clauses.

Mistake 3: Confusing "session usage" with "monthly usage"

A 5-hour session can consume 15% of a weekly quota — that's what Kahn experienced. If your usage consists of long, infrequent sessions, the per-session promise may work in your favor. If you work in short sessions all day long, it structurally disadvantages you. The solution: model your own usage pattern before choosing a plan.

Mistake 4: Ignoring that the "fine print" may be legally insufficient

The exact information existed at Anthropic — two clicks away. The judge considered that wasn't enough. But as a consumer, don't count on a court to protect you: the only effective protection is reading the limits documentation before purchase, not after your first "quota reached" message.

❓ Frequently Asked Questions

Did the judge say Anthropic lied?

No. He ruled that the advertising was "plausibly misleading" to a reasonable consumer. This is a procedural step: the case can proceed to trial, but liability is not established. Anthropic had not commented on the ruling at the time of the order.

How much real usage did the Max 20x plan actually offer?

According to the plaintiffs' calculations detailed in the order, the $200/month plan provided roughly 6x the weekly usage of the Pro plan, and the $100/month plan roughly 3.5x — versus promised multipliers of 20x and 5x, respectively. The gap stems from the fact that the multiplier applied to the 5-hour session.

Which claims survived the judge's filtering?

Three: negligent misrepresentation, breach of contract, and Smith's CLRA claim (the court finding that Claude Code can qualify as a "service"). Four claims were dismissed with leave to amend, including those based on California's unfair competition and false advertising laws.

Are weekly caps illegal in themselves?

No, and Judge Tse explicitly rejected the claim about limits being tightened mid-billing-cycle. The legal issue concerns the initial communication of the plan, not the existence of the caps. A lab can change its quotas; what it cannot do, under this logic, is sell the plan with a promise whose unit was misleading.

Does this change anything for ChatGPT or Gemini?

Indirectly, yes. The test laid out by the judge — the unit of the promise matching the billing unit, disclosures within the purchase funnel — applies to any LLM subscription. Competing offerings use equally opaque limits, and this ruling gives future plaintiffs a legal template.

Has the class already been certified?

No. The November 6, 2026 order only rules on the motions to dismiss. Class certification is a separate, later step. If it succeeds, all affected Max subscribers could be included without any individual action.

✅ Conclusion

A "20x" that's really worth six, a disclosure two clicks away from the buy button, caps added after the sale: the Claude Max affair shows that AI subscription marketing has just run into its first serious legal limit. Whether or not Anthropic settles, the message to labs is clear — put numbers on your promises in the unit your customers pay in. And the message to users is just as clear: read the limits before you pay, and compare models on what they do, not on what their marketing multiplies.