Anthropic is facing a proposed class action accusing the company of marketing its premium Claude Max plans with usage multiples that do not match what subscribers actually receive. The complaint, docketed June 14 in the U.S. District Court for the Northern District of California, lands the same day Anthropic's new metered credit caps for automated "agent" usage take effect, sharpening a clash between flat-rate AI subscriptions and the metered cost of running today's heaviest workloads.
The Allegations: Max 20x Delivers Six to Eight Times Pro, Not Twenty
The named plaintiff, Washington, D.C. resident Karl Kahn, says Anthropic's marketing materially misled buyers. According to the filed complaint, the $200-per-month Max 20x plan "delivers far less than twenty times the usage of the Pro tier, delivering just six to eight times the usage of Pro," and the $100-per-month Max 5x plan delivers "just three-and-a-half times" Pro rather than five. The complaint also alleges Anthropic "entices consumers to pay the $200 monthly Max 20x subscription by falsely claiming it offers a 50% savings." Kahn subscribed to Claude Pro in June 2025, upgraded to Max 5x in January, then moved to Max 20x in April to support heavy coding, and says a single five-hour programming session at one point consumed 15 percent of his entire weekly allowance.
Metered Credit Caps for Agents Take Effect the Same Day
As of June 15, the Claude Agent SDK, headless command-line runs, the Claude Code GitHub Actions integration, and third-party apps that authenticate through a Claude subscription draw from a separate monthly credit sized at $20 for Pro, $100 for Max 5x, and $200 for Max 20x, billed at standard pay-as-you-go rates. Interactive chat in the browser, desktop app, and Claude Code in a terminal remain on the regular subscription limits. The change is the settled outcome of a year of churn: Anthropic briefly blocked subscription tokens from third-party tools in January, reversed an April ban on agent tools within 24 hours after developer backlash, then landed in mid-May on the metered-credit compromise that now governs every Claude subscription.
The Five-Hour Window Meets a Weekly Ceiling
The measurement problem sits at the core of the suit. Anthropic's plans are governed by tokens, with limits that trigger when a user crosses a threshold rather than after a fixed amount of time, and Anthropic does not publish a precise per-session token budget for any tier. On top of that, a rolling five-hour reset runs in parallel with a seven-day weekly ceiling, so clearing the five-hour reset does not free a user from the weekly cap. The two layers together help explain how a single coding stretch can erase a sizable share of a Max 20x weekly allowance and leave a subscriber buying additional usage on top of a plan they already pay $200 a month for.
Flat-Rate Subscriptions Collide With Agentic Compute
The lawsuit and the new credit caps are two faces of the same pressure. Anthropic now says Claude writes more than 80 percent of its merged code, and as agents work closer to continuously, their compute draw stops resembling intermittent human usage. Flat monthly fees were never priced for always-on workloads, which is why Anthropic keeps tightening limits and why a power user can feel shortchanged at the same time. Rivals OpenAI, Google, and Perplexity all sell $100-to-$200 monthly tiers on the same usage-limit promise, which means the outcome here could reset expectations across the category. Anthropic, which confidentially filed for IPO this month, and OpenAI, which has its own confidential S-1 on file, both face investor pressure to close the gap between subscription revenue and inference cost. The complaint brings claims under California's Consumers Legal Remedies Act and False Advertising Law, plus negligent misrepresentation and breach of contract, and seeks damages, restitution, and injunctive relief on behalf of all U.S. residents who bought or upgraded to Max 5x or Max 20x between April 9, 2025 and the present. Anthropic announced the new agent credit model earlier this month; the company declined to comment on the suit.
Reporting based on coverage from Tech Times, Engadget, and the U.S. District Court for the Northern District of California.