by /
Your AI tools may charge different accounts for the work they do. This page helps your team trace each job to the bill it uses. Start by checking your own usage screen before you compare costs.
This guide is part of How We Build and Maintain AI Agents. Next, explore Cursor and Local Qwen Are Not the Same Qwen, or A Skill Is How. A Routine Is When..
Tool billing comparison. Use this page to identify which account pays for the work. For the local worker setup, follow our local Qwen installation guide.
You are not buying one brain. You are buying four cafeterias. The model names look the same. The lunch tickets do not.
MonthlyCursor included pools reset
WeeklyClaude, Codex, GrokBot
Separatefour meters, four piggy banks
WatchGrok Bot weekly bar before routines
This page is the team SOP for where the money actually goes when you run agents. What Grok Bot actually added (a named roster, not Heavy’s four-at-once reply) lives at /grok-bot/. Claude scheduled task = Grok Bot routine = Cursor Automation — skill is how, routine is when. This page sits next to the skill vs agent anatomy, the job that runs it, and the overnight worker that takes batch writing off the buffet entirely. A skill is a document. A harness is files plus a schedule. A lunch ticket is what pays the model that reads those files.
Standing rule. The four meters are separate. Cursor Models, Other Models, Grok Bot weekly, and grok.com Extra Credits do not pour into each other. Watch the Grok Bot weekly bar before you add routines. Open your own spending tab — do not copy someone else’s snapshot. Glossary: /skills-and-routines/.
Prices below are the public plan grids as of 18 Aug 2026. Vendors change them. Check your own dashboard before you annualize a scary day.
The four piggy banks
These four numbers are different piggy banks:
- $300/month for SuperGrok Heavy on grok.com. That is grok.com’s weekly buffet, and one door into Grok Bot. It is not the named-Bot product by itself.
- $200/month for Cursor Ultra. That is the workshop, and the other door into Grok Bot. You do not need grok.com if you live in Cursor.
- $400 on the Ultra row, which looks like extra Grok.
- 2× included usage at half price for Grok 4.6’s first week in Cursor.
Those are still four different piggy banks for usage. The $400 is not extra Grok. The launch week is a sale, not the forever rate. Linking SuperGrok Heavy can unlock Cursor Ultra at $0. The extra $100 on Heavy versus Ultra is grok.com’s top Chat / Imagine / Voice / Build buffet and the hard-problem mode — not a bigger Cursor, not a bigger Grok Bot. The meters still do not pour into each other. Product split: Grok Bot desk SOP.
They are harmonized — and Cursor is the odd one out
Claude, ChatGPT, and grok.com now rhyme. Each vendor sells a chat booth and a workshop that share one allowance. When the allowance is gone, you buy extra at grocery (API) prices.
Cursor is not grok.com’s workshop. Cursor is a fourth cafeteria that rents Grok, Claude, and GPT. Grok Bot (the named roster, announced 11 Aug 2026) is a different product from grok.com chat and from Heavy’s four-at-once reply. Ultra includes Grok Bot. Heavy includes Grok Bot and can unlock Ultra at $0 if you link. The usage bars still do not pour into each other. Same model family, different cash register. Desk SOP: /grok-bot/.
How the windows reset
Think of two clocks on Claude and Codex, and one bigger calendar on Cursor.
| Clock | Who | What it means |
|---|---|---|
| 5-hour burst | Claude (claude.ai, Code, Cowork). Codex local + cloud tasks. | A rolling “slow down” window. You can keep working after it fills, but not at the same rate, until five hours from the start of that session pass. |
| ~7-day weekly cap | Claude all-models weekly. Codex extra weekly limits. GrokBot one shared weekly pool. | The real ceiling. Hit it and the buffet closes until your assigned reset. Claude’s reset day is fixed on your account, not “seven days from when you started.” |
| Monthly billing cycle | Cursor both pools | Resets on your billing date. Unused does not roll over. This is why a week of Cursor Grok 4.6 can feel bigger than a week of Cowork. |
What the $20 / $100 / $200 / $300 tickets actually buy
| Ticket | List price | Included work | After it runs out |
|---|---|---|---|
| Claude Pro | $20/mo | Baseline 5-hour + weekly (shared with Code / Cowork) | Wait, or usage credits at Claude API rates |
| Claude Max 5x | $100/mo | About 5× Pro per session, plus a weekly all-models cap | Same extra-credit cliff |
| Claude Max 20x | $200/mo | About 20× Pro per session, same weekly-cap shape | This is how a power user still burns thousands: the $200 is the buffet, not a ceiling |
| ChatGPT Plus | About $20/mo | Codex “a few focused sessions a week”; 5-hour message ranges (example: GPT-5.6 Sol 15–90 local messages / 5h) | Codex credits (Plus/Pro). Optional auto top-up |
| ChatGPT Pro | 5x or 20x Plus | Same 5-hour window, much larger ranges (Pro 20x Sol 300–1,800 local / 5h). Weekly limits may still apply | Same credit cliff |
| SuperGrok | $30/mo | One weekly pool across Chat, Imagine, Voice, Build. Does not include Grok Bot. | Extra Usage Credits from $5, expire in 1 year, standard rates (worse than the buffet) |
| SuperGrok Heavy | $300/mo | Largest grok.com weekly pool. Door into Grok Bot. Linking can unlock Cursor Ultra at $0. Usage still does not pour into Cursor Models | Same Extra Credits on grok.com. A long agent day is how you see $300 in 24 hours on top of the $300 plan |
| Cursor Pro | $20/mo | Generous Cursor Models pool (Grok 4.6 / 4.5 / Composer 2.5) + $20 Other Models | On-demand at API rates if you enable it |
| Cursor Pro Plus | $60/mo | Same Cursor Models pool + $70 Other Models | Same on-demand switch |
| Cursor Ultra | $200/mo (or $0 if Heavy is linked) | Same Cursor Models pool + $400 Other Models. Includes Grok Bot. Pro / Pro+ do not. | The $400 is Claude/GPT/Gemini inside Cursor, not extra Grok Bot, not a bigger Grok 4.6 pool. Grok Bot usage is a separate weekly meter |
Cursor Models size is described as “generous,” not a public dollar figure. ChatGPT Plus/Pro dollar rows follow OpenAI’s public Codex grid (5x / 20x of Plus) rather than a screenshot of one invoice. Claude Max prices: Anthropic help, 18 Aug 2026. SuperGrok Heavy: grok.com plans / xAI FAQ.
Cursor’s $400 is the other piggy bank
Ultra’s $400 is the Other Models bank, not extra Grok.
Stay on Cursor Grok 4.6 and you spend the teal bank. Pick Claude or GPT inside Cursor and you spend the orange bank — $20 on Pro, $70 on Pro Plus, $400 on Ultra — at that model’s API price. Ultra did not buy you $400 of Grok. Ultra bought you a bigger Claude-inside-Cursor wallet, plus the same generous Grok/Composer pool every paid plan already gets.
On-demand list prices if that teal box ever empties and you turn the extra switch on (per million tokens): Grok 4.6 $2 in / $0.50 cached / $6 out. Fast is double. Composer 2.5 is cheaper ($0.50 / $2.50). Claude Opus 5 inside Cursor is $5 / $25. Claude Fable 5 is about $10 / $50. That is why picking Claude in Cursor for a 45-article night can feel like Cowork again.
This week is on sale. Do not annualize it.
Grok 4.6 launched 12 Aug 2026. Cursor and Grok Build advertised 2× included usage for the first week. Cursor also listed a 50% on-demand discount for that same week. Today is still inside that window. A dashboard that says 4% used is 4% of a temporarily doubled buffet, after a day of work that would have been extra credits on Claude or GrokBot.
After about 19 Aug 2026 the 2× and the 50% go away. The monthly-vs-weekly shape stays. The two piggy banks stay. The “crazy cheap” feeling will shrink. Plan the factory on the normal pools, not on launch week.
How extra usage gets charged (why a task has no price tag)
A subscription ticket is a buffet. The kitchen does not print a menu on your plate. You see a percentage, a progress bar, or “you have extra credits left.” That is not the same as the API invoice, where every million tokens has a published dollar rate.
When the buffet closes, every vendor switches you to something that is priced like the API, even if they still call it “credits”:
| Shop | What they call extra | How it actually bills |
|---|---|---|
| Claude | Usage credits (optional bundles) | Standard API rates after the included 5-hour / weekly limits. A long Fable Cowork thread is Opus-class money. |
| ChatGPT / Codex | Credits + auto top-up | Included first, then the credit balance. OpenAI says Codex extra is token-based (input / cache / output), not a flat “per task” stamp. |
| Grok Bot (named roster) | Weekly included usage, then Extra Credits / on-demand depending how you signed in | Cursor: Grok Bot usage resets weekly, separate from Cursor Models. grok.com Extra Credits (if you are on Heavy chat/Build) are still grocery prices, expire in a year. Do not mix the new roster with grok.com chat. |
| Cursor | On-demand usage | Other Models: published API prices, so you can do the math. Cursor Models: the included pool is a %, on-demand uses the Grok 4.6 / Composer table if you enable it. |
Why nobody can tell you “this task costs 12 credits” up front:
- A task is not a unit. Tokens in + tokens out + thinking tokens + tool calls + the previous files stuffed back into context are the unit.
- Higher effort (Claude, Cursor Grok 4.6 xhigh) buys more thinking. Same prompt, more tokens.
- Fast on Cursor Grok 4.6 is 2× list price and is the default speed on Pro and up.
- A 500-step Cowork thread re-reads the whole history on every correction. That is the S-curve we already wrote about on /claude/.
- Cursor Grok 4.6 has an effort picker: low / medium / high (default) / xhigh. grok.com chat often does not. Grok Bot is a different product from that chat booth — see /grok-bot/.
The honest meter is the dashboard after the run: Cursor Spending, Claude Settings → Usage, grok.com Settings → Usage, Codex Settings → Usage. If extra credits or on-demand are on with no cap, the red box has a blank check.
The meters stay separate
Cursor Models, Other Models, Grok Bot weekly, and grok.com Extra Credits are four piggy banks. A quiet Cursor bar after a heavy factory day does not mean Grok Bot is quiet. Extra credits after a weekly ticket are grocery prices. Do not annualize one extra-credit day as if it were Cursor. Heavy four-at-once replies and named Grok Bots are not the same product. Watch the Grok Bot weekly bar before you add routines.
Where the work should live (this is the harness)
The System already says: context in files, skills as documents, agents as roles, jobs as schedules. Money routing is the same ladder with a price on each rung.
| Work | Put it here | Why |
|---|---|---|
| Write article 17 of 45, scripts, fleet chores | Cursor Grok 4.6 (teal bank). Effort medium/high. Fast off if the month is stretching. | Monthly included pool + tools on disk. Do not pick Claude inside Cursor for this. |
| 790 leftover Sean Kelly drafts overnight | Local Qwen on the Mac | Electricity. No vendor ticket at all. |
| Honest score, real voice, consequential call | Thin Claude slice if Grok’s draft is wrong | That is the only work that earns Max/Fable prices. Keep it short so it stays inside the weekly buffet. |
| Named always-on roles: inbox, CRM, websites with no API, phone handoff | Grok Bot (the roster). Not grok.com chat, not Heavy’s four-at-once reply. | That is the new product: named staff on a shared cloud computer. Cursor still owns the files. |
| Click wp-admin, git, fleet, skill files, factory | Cursor agent (Codex cousin). Claude Chrome only when you mean to spend Claude. | Building lives in the workshop on disk. Skip grok.com if you need a harness. |
If handoff feels heavier than paying extra credits, the missing piece is still the job (schedule, QA, working files), not a smarter model. Subscriptions buy execution; the shared records, checkpoints, and receipts stop each new chat from relearning the work. See the canonical guide to agent memory and coordination.
Mechanic checklist. 1) Pick the shop (Cursor for Process). 2) Stay on Cursor Models unless you mean to spend Other Models. 3) Effort high, not xhigh, unless you are stuck. 4) Fast off when stretching the month. 5) Extra credits / on-demand off, or capped. 6) Overnight local for bulk drafts. 7) Look at the meter weekly, not after a $300 surprise.
Check your own meter
- Cursor: cursor.com/dashboard/spending — both pools, reset date, on-demand.
- Claude: Settings → Usage — 5-hour bar, weekly bar, extra credits.
- Grok Bot: weekly usage on the Grok Bot / Cursor Grok Bot meter. grok.com Settings → Usage only if you are actually on grok.com chat/Build Extra Credits.
- Codex: Codex Settings → Usage — 5-hour ranges, credits, auto top-up.
Where this sits
A skill is a document. An agent is a role. A job is a schedule. This page is the price of the brain that reads those files.
The System · Skill vs agent · Persistent agents · Skills and routines · Grok Bot desk · Overnight worker · Let the model pick the model · Why we used Claude hard
THE DELIVERABLE
Stay on the buffet. Cap the grocery box. Put bulk writing on the Mac.
Overnight worker Persistent agentsStanding SOP, 18 Aug 2026. Public grids: Cursor models & pricing, Grok Bot plans, Heavy unlocks Ultra, Grok 4.6 launch (2× first week), Claude Max, Claude usage credits, Codex pricing, xAI Grok FAQ, Introducing Grok Bot. Desk: /grok-bot/. Receipt: how we mapped the usage buckets.
Dennis Yu
Dennis Yu is the CEO of Local Service Spotlight, a platform that amplifies the reputations of contractors and local service businesses using the Content Factory process. He is a former search engine engineer who has spent a billion dollars on Google and Facebook ads for Nike, Quiznos, Ashley Furniture, Red Bull, State Farm, and other brands.
Dennis has achieved 25% of his goal of creating a million digital marketing jobs by partnering with universities, professional organizations, and agencies. Through Local Service Spotlight, he teaches the Dollar a Day strategy and Content Factory training to help local service businesses enhance their existing local reputation and make the phone ring.
Dennis coaches young adult agency owners serving plumbers, AC technicians, landscapers, roofers, electricians, and believes there should be a standard in measuring local marketing efforts, much like doctors and plumbers must be certified. He has appeared on 353 podcasts with 619 credited episodes — see the full list of his podcast appearances.