Today's Thread
AI subscription prices are holding steady, but the way usage is counted is changing. Anthropic removed the five-hour limit from Claude's top plan, and OpenAI is merging chat and coding-tool usage into a single pool. Both changes shift the question from how much you pay each month to what drains how much inside that payment.
One of those drains came to light. A measurement found Claude Code spending more on a 3–5 word status line than on the main conversation, and tools like herdr, which let you watch several coding AIs in one window, are widely used. The more AI workers you add, the more you need to track how many run at once and where usage goes, not just which plan you are on.
Claude Max 20x becomes Ultra — five-hour limit dropped
On October 30, Anthropic renamed its paid Claude plans. The $100-a-month Max 5x is now Max, the $200-a-month Max 20x is now Ultra, and prices are unchanged. According to the official pricing page, Ultra allows 10 times Pro's usage per week with no five-hour reset limit for verified accounts, while Max keeps 5 times Pro for both the five-hour and weekly windows. Both plans include API credits, the money used when your own programs or automations call Claude: $100 a month on Max and $200 a month on Ultra.
The old name 20x read like 20 times Pro, but the official help article lists Ultra's weekly usage as 10 times Pro, and the weekly limit itself did not change this time. What actually changed is the five-hour window. Before, heavy bursts meant waiting until the next five-hour window opened; verified Ultra accounts are no longer bound by time within the weekly limit. Anthropic verifies accounts automatically from account and payment history, separate from the phone-number check at checkout.
This helps people who batch their work, such as revising dozens of documents in one day or running long coding sessions. Since the weekly total is the same, people who use a little every day will notice little difference. When a team chooses between Max and Ultra, the first question is whether they work in bursts or spread usage out, not the price.
Anthropic has not disclosed how long automatic verification takes or what criteria it uses. Unverified accounts keep a temporary five-hour limit and receive an email when it is lifted. Ultra is billed monthly only, and mobile app pricing may differ from the web.
Claude Code's 3–5 word status line: 16.4% of a week's spend
According to a measurement published in late October by outside developer Flamerix AI, Claude Code resends each background agent's full conversation to the model about every 30 seconds to produce a 3–5 word status line. An agent is an AI worker that carries out several steps on its own, and background means it runs behind the scenes, separate from the conversation you see. Over the measured week, these summary calls made up 16.4% of Claude Code's own cost estimate, more than the main conversation's 15.7%, and 19.7% of requests. The feature has been on by default since version 2.1.232, released August 13.
The status line only appears in the agents panel and adds nothing to the work itself. Yet there is no setting to turn off just the summaries; blocking them through official settings means disabling background agents altogether. The measurer published a short script that blocks only this timer, but critics noted it must be rechecked after every Claude Code update and is loaded into other programs sharing the same runtime, and advised against it.
The summary runs separately for each background agent. If you keep several agents running at once, this may be one reason your limit drains faster than expected, and running fewer at a time cuts this cost too. Teams need usage records broken down by request type to spot leaks like this.
The 16.4% figure comes from one team's week of work and varies widely by workflow. Anthropic's official figures or any plan for a setting to turn off only the summaries cannot yet be found in its official documentation. The workaround script is not an official feature, so on a work computer it is safer to consult your security team first.
ChatGPT and Codex will draw from one usage allowance
In an email to subscribers, OpenAI said ChatGPT and Codex usage will draw from the same allowance. Codex is OpenAI's coding agent that writes and fixes code for you. According to reports, the change covers Plus, Pro and Business plans and applies from December, and conversations with Instant, the fast default model, do not count. Remaining usage and reset times will be shown together on the Usage screen in Settings; when you run out, you can wait for the reset, upgrade, or buy credits.
For people who use both chat and Codex, heavy use of one now shrinks what is left for the other. OpenAI explained that more advanced models and complex tasks draw down more. Earlier, on October 29, it doubled Work and Codex usage on Pro 100, 200 and 500, and said a test bringing Work, its business features, into the regular chat screen will start with a small group in the coming weeks.
If you organize material in chat and build automations in Codex, both jobs now share one pool. Sending short questions to Instant, which does not count, and saving advanced models for work that truly needs them is now a form of cost control.
How much each plan gets, whether the current five-hour and weekly windows stay, and the exact start date have not been announced. Write down your usage before the change so you can compare afterward.
herdr runs multiple coding agents in one terminal
herdr is a public tool that runs several coding agents such as Claude Code, Codex, Cursor and OpenCode side by side in one terminal window. A terminal is the black window where you type commands instead of clicking. Each pane shows whether it is working, waiting for your answer, or idle, and work keeps running in the background even if you close the window or lose a remote connection. On a Mac it installs with one line through the Homebrew installer, brew install herdr, and it has passed 47,000 stars on GitHub.
Earlier terminal-splitting tools were for people to divide windows. herdr adds a control channel for agents, so an agent can open new panes, send instructions to another agent, and wait until that agent gets stuck. Agents running on remote machines appear in the same list.
When you split work across several agents, it is easy to miss which window has stopped and is waiting for you. A single status indicator cuts that idle time. As the previous item shows, though, the number of agents is your usage, so decide how many to run at once first.
After a restart the pane layout comes back, but the original running processes do not. The Windows version is still in beta.
Try Today
If you use Claude Ultra, open Settings > Usage and check for a Current session bar. If there is none, your account is verified and has no five-hour limit. If you use Max, note how many times you hit the five-hour limit this week, and if it happens often, consider moving up to Ultra.
Count how many background agents you run at once in Claude Code. If your limit drains quickly, compare a week with fewer agents against your current usage. The workaround script is not an official feature, so do not install it on a work computer.
Record your remaining usage and reset time on ChatGPT's Usage screen in Settings. It becomes your baseline for comparison after December. Start sending short questions to Instant today.
If you use two or more coding agents, install herdr and run just two side by side. Simply seeing which window is waiting for you shows how much time was being lost to stalls.