In June 2026 a Claude Max customer named Karl Kahn filed a proposed class action against Anthropic in federal court in California, alleging that the Max plans deliver far less than their advertised multiple over the cheaper Pro plan (Engadget, PYMNTS).
Start with the honest caveat, because a lot of the coverage skipped it. The suit was just filed. Nothing has been adjudicated. Nobody has proven that Anthropic delivered less than it promised, and a complaint is a set of allegations, not a finding. So this is not a story about a company that "lost." It is a useful prompt to understand something every AI buyer should know before they upgrade: the headline number on a subscription tier is not what governs what you actually get. The rate limits are.
What is actually true here
Set the allegations aside and look at the documented facts.
Anthropic sells Claude on tiers: Pro at $20 a month, Max "5x" at $100, and Max "20x" at $200 (PYMNTS). The names suggest you get five or twenty times the Pro plan's usage.
In late July 2025, Anthropic added weekly rate limits on top of the existing rolling five-hour reset, and said the change would affect "less than 5% of subscribers" (TechCrunch, Anthropic on X). In May 2026 the company announced higher usage limits alongside a large compute deal with SpaceX, framing more compute as the path to lifting limits (Anthropic, CNBC).
The complaint alleges the Max 20x plan delivers only about six to eight times Pro, and the 5x plan about three and a half. Those are allegations, not established numbers, so I am not repeating them as fact. Reporting also notes the plaintiff says one five-hour coding session ate about 15 percent of his weekly allowance (Engadget).
You do not need the lawsuit to be right to take the lesson. A multiplier in a plan name is a marketing label. Rolling windows, weekly caps, and peak-hour throttles are what decide your real throughput.
How to read an AI subscription before you buy the big tier
Treat any advertised multiple as a ceiling, not a guarantee, and do this before you upgrade:
- Find the actual rate-limit mechanics in the provider's own docs. Look for the rolling window length, any weekly cap, and whether limits tighten at peak hours. That fine print, not the tier name, is the product.
- Run a one-week measured pilot on the cheapest tier. Use the plan the way you actually work for a week and watch where you land. You will learn your real consumption before you pay quadruple for a number you might never reach.
- Budget by measured throughput, not by the label. Think in sessions per week or hours of model time, the units you observed in the pilot, rather than "20x sounds like plenty."
- Watch for capacity-driven changes. When a provider ties limit increases to a new data-center or compute deal, that is a signal limits can move under you in either direction. Re-check after any such announcement.
- Keep your workflow portable. If caps tighten or pricing shifts, you want to move to another provider without rebuilding everything. I wrote about that cost discipline in the cheaper-models piece linked below.
The bigger signal worth watching
There is a separate development worth knowing, and I want to be careful not to overstate it. In June 2026 a regional court in Munich ruled that Google's AI Overviews are Google's own statements rather than neutral search results, making Google liable for them (The Decoder). That case is about Google, not Anthropic, and it does not bind anyone else. But it points at a direction: courts are starting to treat what AI companies say and show as commitments, not disclaimers. If that holds, the gap between a marketed "20x" and a delivered experience becomes a legal question, not just a community gripe.
For the small-business owner, the takeaway is simpler than any of the legal drama. Buy the tier your measured usage justifies, read the limits before the label, and keep the freedom to leave. That mindset, applied across your whole tool stack, is the spine of my book The $20 Dollar Agency (search the title on Amazon Kindle).
Related reading
- Why Use the Separate Anthropic API, and How It Actually Works: when metered API pricing beats a capped subscription.
- I Cut a Recurring AI Bill by More Than Half in an Afternoon: the routing and caching habits behind paying for what you use.
- Western Tech Is Routing AI Workloads to Chinese Models: the cheaper-model option if your tier stops being worth it.
- How a Small Business Runs AI Agents Without a Surprise Bill: budgeting AI by measured usage.
Fact-check notes and sources
- Kahn v. Anthropic: a proposed class action filed June 2026 in the Northern District of California over Claude Max usage limits; the case is unadjudicated and its claims are allegations (Engadget, PYMNTS).
- Pricing: Pro $20, Max 5x $100, Max 20x $200 per month (PYMNTS).
- Weekly rate limits: introduced late July 2025 on top of the five-hour rolling reset; Anthropic said the change would affect "less than 5% of subscribers" (TechCrunch, Anthropic on X).
- SpaceX compute deal: announced May 2026, explicitly tied to higher usage limits (Anthropic, CNBC).
- Munich ruling on Google AI Overviews: June 2026, holding the overviews are Google's own statements; this concerns Google only (The Decoder).
- I have not repeated the complaint's "six to eight times" allegation as fact, nor the "less than 2%" figure that appeared in some retellings; Anthropic's stated figure was "less than 5%."
This post is informational, not legal or financial advice. The lawsuit described is unresolved and its claims are unproven allegations. Mentions of Anthropic, Claude, Google, and other third parties are nominative fair use. No affiliation is implied.
