Guides › Save money

How to stretch your AI plan limits

Updated Oct 30, 2026.

Hitting your AI limit halfway through the week? Before you pay for a bigger plan, figure out which limit you're hitting, cut the waste, and spread the work across the plans you already have. If you pay for two, you can even hand heavy background jobs to the second one while your main assistant keeps track of them.

Session limits vs weekly limits

Most AI plans now have two meters. A session limit resets after a few hours; a weekly limit resets once a week. Running out of the first costs you an afternoon. Running out of the second costs you days.

App	Short window	Weekly	Where to check
Grok Bot	None listed	A weekly usage pool. On The X Subscription it's shared with Cursor and the Grok apps; prepaid credit can cover usage past it. ³	Usage at grok.com/dashboard (X Subscription) or Settings › Usage & Billing ³
Claude	Resets 5 hours after your first message. Verified Ultra accounts have none. ¹	Resets at a fixed time each week; some plans have a separate weekly limit for Fable. ¹	Settings › Usage ¹
Gemini	Refreshes every 5 hours until you reach your weekly limit ²	Yes ²	Settings › Usage limits ²

Renamed Oct 30, 2026: Max 5x is now Max, and Max 20x is now Ultra, at the same prices. Max 20x was billed as 20x Pro per 5-hour session. Ultra is billed as 10x Pro weekly. Anthropic says weekly limits haven't changed, but it's unclear whether the actual allowance did. Anthropic also removed the 5-hour limit for verified Ultra accounts, starting with most existing subscribers. Newer accounts keep a temporary limit until they're verified. ^4,5,6

That means a verified Ultra subscriber can use a whole week's allowance in one long day, so the weekly meter is now the one to watch. claude.com lists Max at 5x Pro usage every 5 hours and every week, and Ultra at 10x Pro usage every week. ⁵

What uses up a limit fastest?

Long chats. Every message re-sends the whole conversation. ^7,2
The biggest model or highest effort on routine work. Anthropic says Opus uses meaningfully more of your quota than Sonnet. ⁷
Deep Research, images and video. Google lists these among the heaviest uses. ²
Connected tools you don't need in that chat. ¹

Spread work across plans by strength

If you already pay for more than one plan, stop using one app for everything. Give each job to the plan that does it best, and each allowance lasts longer.

Grok Bot: your coordinator. It works from a cloud computer with a browser and a command line, and keeps going with your laptop closed. Let it plan, keep track and run routines. ^8,3
Claude: long coding and writing jobs. Claude Code comes with every paid Claude plan and shares the plan's limits. ⁵
Gemini: very long documents. Google AI Pro and Ultra read up to 1 million tokens at once, about 1,500 pages. ²
ChatGPT: if you have it, its agent mode (Work) and Codex share one allowance, so a big Codex job leaves less for everything else. ⁹

Prompt: plan my week across my AI plans

I pay for these AI plans: [list them].
Here's what I need to get done this week: [list].
Assign each job to the plan best suited for it, flag the heavy ones, and spread them so I don't hit any one plan's weekly limit. Show it as a table: job, plan, why, rough size (small/medium/heavy).

Route heavy background jobs to a second plan (step by step)

The idea is simple. Your main assistant stays the coordinator. When a long, self-contained job comes up, it hands the job to Claude Code running on your own Claude plan, waits, and reports back when it's done. The heavy work draws from your Claude allowance instead of your main assistant's.

Fair warning: one widely shared X post this week claims this makes limits "last the whole week". We haven't tested that, and it depends on your jobs. Claude Code still counts against your Claude plan's 5-hour and weekly limits, the same pool as the Claude app. ¹ We don't recommend any ready-made bot or template for this. The steps below use only official tools.

Pick the right job. Good fits are long, self-contained and checkable: a code refactor, writing tests, tidying a folder of docs, a big review. Quick questions aren't worth routing.
Install Claude Code where the job will run. That's your assistant's cloud computer (Grok Bot has a command line) or your own computer. Anthropic's installer for macOS and Linux is curl -fsSL https://claude.ai/install.sh | bash, then check it with claude --version. ^10,8
Sign in once with your Claude plan. Run claude and finish the login in your browser yourself. Pro, Max and Ultra all work, and Claude Code remembers the login. ¹⁰ On Grok Bot's computer, command-line logins are shared by all your Bots, so only do this if that's OK with you. ⁸
Write the job down. Put it in a file such as JOB.md: the goal, which files to touch, what "done" looks like, and what not to do.
Have your assistant start it in the background with Claude Code's non-interactive mode, claude -p. It exits with code 0 on success. ¹¹
Have your assistant report back. It watches for the job to finish, reads the result, and sends you a short summary: what changed, what failed, what needs your call.
Review before anything ships. Check the changes yourself, just as you would a colleague's work.
Make it a habit once it works: save the steps as a skill, or set a routine to run it on a schedule. ³

Show me the command

Run from the project folder. --permission-mode acceptEdits lets Claude edit files without asking; shell commands still need an --allowedTools entry. --output-format json puts the answer in a result field. ¹¹

claude -p "$(cat JOB.md)" --permission-mode acceptEdits --output-format json > result.json; echo "exit: $?" >> job.log

Don't add --bare: bare mode skips your subscription login and needs an API key, which bills separately. ¹¹

Prompt: hand a heavy job to Claude Code

You're my coordinator. This job is heavy, so don't do it yourself: [describe the job].
1. Write it up in JOB.md in [project folder]: goal, files to touch, what done looks like, and what not to change.
2. Start it in the background with: claude -p "$(cat JOB.md)" --permission-mode acceptEdits --output-format json > result.json
3. Check on it until it finishes. Don't edit the files yourself while it runs.
4. When it's done, tell me: did it succeed, what changed (list files), anything that failed, and anything that needs my decision. Don't commit, publish or send anything until I say so.

What can go wrong?

The second plan runs out too. Claude Code shares your Claude plan's 5-hour and weekly limits. When you hit one, you wait for the reset or turn on usage credits. ¹
It stops to ask permission. In -p mode, anything that needs approval is denied unless you allowed it up front, so list the tools the job needs. ¹¹
It wanders. A vague job wastes tokens. Ask for a plan first on anything touching more than a few files. ⁷

When to upgrade instead

You hit the weekly limit most weeks even with good habits. Moving up a tier may cost less than juggling plans.
You only run out at the end of the week or once a month. Claude's paid plans let you turn on usage credits so work continues at API rates instead of stopping. ^1,5
You have usage to spare before your reset? Spend it: 15 high-value ways to use leftover AI usage.

Compare plans: Claude plans and prices · Cheapest good AI (US and Canada) · Get more out of your Claude plan for less.

Common questions

4 quick questions, answered

Does routing jobs to a second plan really make my limits last all week?

It can help, because the heavy job draws from the second plan's allowance instead of your main assistant's. But nobody can promise a whole week: it depends on how big your jobs are, and the second plan has its own 5-hour and weekly limits. We haven't tested the "lasts all week" claim.

Does Claude Code use the same limits as the Claude app?

Yes. Anthropic says claude.ai, Claude Code and Claude Desktop all count toward the same usage limit, so a big Claude Code job uses the same allowance as your chats.

What changed with Claude Max on Oct 30, 2026?

Anthropic renamed Max 5x to Max (still US$100 a month) and Max 20x to Ultra (still US$200). It removed the 5-hour limit for verified Ultra accounts, starting with most existing subscribers; newer accounts keep a temporary limit until they're verified. Anthropic says weekly limits haven't changed, but it's unclear whether the actual allowance did.

Should I buy a second plan just to do this?

Usually not. Route work to plans you already pay for. If you'd need a new plan, compare it with simply upgrading your main one.

^6,5,1.

Sources (11)

Last checked Oct 30, 2026.