Claude Usage Limits: A Lawsuit Says 20x Doesn’t Buy You 20x

TL;DR: A class action filed on 8 September alleges that Anthropic’s “5x or 20x more usage” claim for Claude Max describes only the five-hour session window — while a separate weekly cap, added months after Max launched, is the one most heavy users actually hit. Claude usage limits, it turns out, come in two layers. Anthropic disputes this, and says the limits are published. Nothing here is proven. But the underlying question is a fair one for anyone paying $200 a month: what does that actually buy you?

What the lawsuit says

Anthropic sells Claude Max in two tiers — $100 for “5x” and $200 for “20x” more usage than the Pro plan, which starts at around $17 a month billed annually.

A class action filed on 8 September, first reported by The Verge, alleges those multipliers describe usage within a five-hour session window that resets, and that Max subscribers are also subject to a weekly cap — one introduced in the summer of 2025, months after Max launched in April of that year.

It follows an earlier federal suit brought in June by a single subscriber, first reported by the Wall Street Journal. That complaint alleges “the actual usage provided by the Max 5x and Max 20x plans is far below the advertised amount”. The plaintiff says that after upgrading to Max 20x for Claude Code work, one five-hour session consumed 15% of his weekly allowance.

Everything above is an allegation. None of it has been tested in court, and Anthropic disputes the characterisation.

How Claude usage limits actually stack

Setting the lawsuit aside, the mechanics are not in dispute and are worth understanding on their own:

Limit	How it works
Session window	Rolling five hours; this is what the 5x / 20x multipliers describe
Weekly cap	A separate ceiling across all models, applied on top
Further limits	Anthropic says it may add more restrictions to manage capacity

Claude usage limits come as two ceilings, not one, and the advertised number describes the first. For light use you’ll never meet the second — and for anyone running Claude Code hard, the second is the one that stops you.

There’s a widely-shared calculation doing the rounds that illustrates the gap: buying the $200 tier over the $100 one looks like it should give you four times the usage, but once the weekly cap is applied it reportedly works out closer to 1.7 times, for twice the price. Worth flagging clearly — that figure comes from a user’s own calculation in a viral Reddit post, reported by Engadget. It is not a court finding and not Anthropic’s number.

Anthropic’s side

Which deserves setting out properly, because the company has an argument.

Anthropic does publish the limits on a Claude support page. In a motion to dismiss the earlier version of the case, per The New Stack, it argued customers could reach that information through hyperlinks during purchase — comparing it to a product label you can turn over before deciding to buy.

The plaintiffs’ counsel, Monica Vaca, puts the counter-argument: the terms require clicking through several hyperlinks, and “this is hard for consumers — they don’t know what’s in the black box.”

There’s also a reason the weekly cap exists that isn’t sinister. Anthropic introduced it in response to people running Claude Code “continuously in the background, 24/7” — a genuine capacity problem, and one that pure session limits don’t solve.

Anthropic declined to comment on the earlier suit, and had not responded to Engadget or The New Stack on the new one at the time of their publication.

Why it matters if you don’t code

Here’s the thing that makes this our story rather than a legal one.

A multiplier is a terrible unit for people who can’t estimate the denominator. A developer has a rough sense of what a heavy session consumes. If you’re building an app as a means to an end, you have no intuition at all for how much usage a task costs — and the costs are wildly uneven. As one expert quoted by The New Stack put it, a single autonomous debugging loop can burn millions of tokens through context re-submission and tool calls. Two intense afternoons can eat a week.

So “20x” answers a question you weren’t asking. You want to know how much work can I get done for £X. The multiplier only tells you how one plan compares with another plan you also can’t measure.

Three consequences worth holding onto:

• Hitting a wall isn’t your fault, and it isn’t necessarily a fault. Two ceilings exist; the advertised number is the smaller one.
• Of the two Claude usage limits, the weekly cap is the one that matters — and it’s the one that isn’t in the marketing.
• You may not know why your usage vanished. Which is uncomfortably relevant this week: we wrote yesterday about attackers stealing Claude sessions and spending people’s allowances. Same symptom, entirely different cause — and no itemised usage history to tell them apart.

How to work out what you’re actually getting

Since no plan page will tell you, measure it yourself. It takes a week and it beats any multiplier.

1. Note your weekly usage percentage at the start of a normal working session, and again at the end. That single number tells you more than “20x” ever will.
2. **Do it on a typical task, not your biggest one.** You want your ordinary week, not your worst day.
3. Multiply up. If a two-hour session costs 8% of your week, you have roughly 25 hours of that kind of work. Now you can judge £200 honestly.
4. Check the weekly figure, not just the session one. The five-hour bar refilling tells you nothing about the ceiling above it.
5. Compare against the cheaper tier before upgrading. If the step up is priced at 2x, satisfy yourself you’re getting near 2x weekly — the disputed claim in this case is precisely that you may not be.

Who should care (and who shouldn’t)

• Paying $200 for Max 20x and using Claude Code hard: the group this is actually about. Run the week-long measurement.
• On Max 5x wondering whether to upgrade: the most useful thing you can do is measure before you spend, not after.
• On Pro at $17–20: you’re unlikely to be near the weekly cap, and this dispute over Claude usage limits doesn’t concern your tier.
• Building on Replit, Lovable or Cursor: not this lawsuit, same lesson. We wrote about what Replit’s routing actually does to your costs — every one of these tools sells capacity you can’t easily measure.
• Curious how it ends: it may not end soon. This is early-stage litigation, and Anthropic is contesting it.

Our take

We want to be careful here, because “AI company sued” is an easy headline and this is not a simple story.

On the narrow legal question we have no view, and neither should you yet — a complaint is one side’s account, Anthropic has a real argument that the limits are published, and no court has found anything. Anyone telling you Anthropic has been caught doing something is ahead of the evidence.

On the substance, though, the plaintiffs are pointing at something true regardless of how the case goes: a multiplier is not a measure of work. “20x the usage of a plan you also can’t quantify” is a comparison, not information, and it becomes genuinely misleading when a second, unadvertised ceiling is the one you actually meet. That isn’t unique to Anthropic — the same piece notes OpenAI’s Codex allowances vary by task, model and where the work runs, and that its Astra model costs 2.5x more per token than GPT-5.6, so identical allowances buy very different amounts of work.

The honest summary is that nobody selling Claude usage limits — or anyone else’s — can currently tell you how much work your subscription buys, because they don’t know either — it depends on what you ask for. Some are experimenting with charging per completed task instead. Until that settles, the only reliable number is the one you measure yourself, and it takes about a week to get.

Not sure which tool — or which tier — fits what you’re building? Take the 60-second Vibe Coding Tool Finder quiz →

FAQ