Anthropic can't stop winning

Summary

Anthropic renamed its $200/month Max 20X subscription to Ultra, keeping the same price and broadly the same offering. The main change is the removal of the five-hour usage limit for verified Ultra accounts. Newer accounts may temporarily retain the limit while Anthropic verifies them.

Subscription plans and limits

Claude’s plans are Free, Pro ($20/month), Max ($100/month), and Ultra ($200/month). Subscriptions have two kinds of usage limits: a weekly limit and a five-hour limit, which prevents users from consuming too much of their weekly allowance in a short burst.

The presenter says the old “5X” and “20X” labels referred to five-hour usage, not total usage. Max 20X’s weekly limit was actually set at 10 times Pro’s. Anthropic’s head of subscriptions acknowledged that the naming was misleading and apologized. The new labels and plan pages are intended to make the limits clearer.

The presenter estimates Ultra’s subscription usage could be worth around $12,000 per month at API rates. He says the five-hour caps help Anthropic manage compute availability during peak demand, rather than simply limiting a user’s spending. Subscription usage can also introduce users to Anthropic’s models, some of whom may later use them through paid enterprise APIs.

Verification and compute availability

Anthropic says most existing subscribers will have the Ultra limit removed, while newer accounts may have a temporary cap to help prevent fraud and abuse. The presenter criticizes the lack of detail about how verification works. According to Anthropic’s guidance, phone verification at checkout is separate. Users can check Settings → Usage: if there is no current-session bar, the account is verified.

The presenter connects changes to compute availability with Anthropic’s access to additional GPUs through a reported SpaceX compute deal. He says that added capacity previously enabled more generous Claude Code limits and the removal of peak-hour reductions. Usage patterns vary substantially throughout the day, so limiting how much subscribers can use in a five-hour window helps reserve capacity for enterprise customers during busy periods.

Impact on T3 Code and agent SDKs

A major issue for users of third-party coding tools was that apps using Anthropic’s Agent SDK appeared to consume the five-hour allowance faster than Claude Code’s interactive CLI. The presenter cites reports and his own calculations suggesting T3 Code used about 1.8 times more of the five-hour window per token. He says OpenCode also uses the Agent SDK.

At the time of recording, the presenter said he had obtained permission from Anthropic to merge a T3 Code change that identifies the tool’s entry point as T3 Code rather than the SDK. He expected this change to make T3 Code users’ five-hour usage comparable to Claude Code’s, and said the fix would ship in an upcoming release. Ultra users would no longer face the five-hour cap once verified, but the issue mattered to users on the $20 and $100 plans.

Sponsored product segment

The video also features Greptile, an AI code-review platform. Its advertised features include review agents that can run code in a sandbox, an editable knowledge base that stays synchronized with a codebase and is available to agents through MCP, and multiple review tiers for different levels of review effort.

Sources and speakers

• Theo (T3.gg): Presenter and source of the analysis, usage observations, and T3 Code update.
• Mike, Anthropic’s head of Claude subscriptions: Quoted on the history of the plan names and limits, including an apology for the confusing “20X” label.
• Anthropic: Source of the subscription and verification information discussed in the video.
• T3 Code users and analytics: Cited regarding faster five-hour-limit consumption in SDK-based tools.
• Greptile: Sponsor featured in the video.