Grok Bot vs Meta Muse vs OpenAI Dots: Which Always-On AI Agent Is Worth Paying For?

Since 11 August 2026, three companies have each launched an AI agent with its own cloud computer that keeps working after you close the app. Grok Bot shipped in beta on 11 August 2026, Muse from Meta launched on 8 September, and OpenAI dots arrived on 29 September. The launches came 28 days and 21 days apart, and they sound identical in the press releases. The price pages, the permission models and the places each one meets you are not identical at all.

The short version: the cheapest way in depends on what you already pay for. Muse is free to start. Grok Bot rides inside Cursor Pro or SuperGrok. A dot comes with ChatGPT Pro or Business Premium. For a four-person team choosing which to try first, the real question is which subscription you already hold, and the table further down has every price.

Who should pick which agent

Start here, then check the numbers below. Each pick follows from a plan you may already pay for.

• A consumer in the US or Canada who lives in WhatsApp: Muse. It is free to start and takes errands.
• A team on ChatGPT Business Premium or Pro that lives in Slack or Teams: a dot. Only dots list those two chat tools.
• A developer on Cursor, or anyone with SuperGrok: Grok Bot. It is already in your plan, so a first test costs nothing extra.

No independent test compares how often these three finish a job. Every capability claim below comes from the vendor. Treat the verdicts as a fit guide, not a ranking, and see the AI agents directory for the wider field.

What each agent is

All three share one idea. The agent gets a computer in the cloud, signs into your apps, works while you are away and comes back when it needs a decision. The differences sit in how each vendor packages that idea.

Grok Bot is SpaceXAI's version, in beta since August. xAI describes Bots that sign into tools "like you do", run in parallel, message each other and learn routines by watching you do a task once. It runs on Grok 4.6 from xAI.

Muse is Meta's personal agent. Meta says it runs on a dedicated Muse Secure VM, a persistent Linux computer with a full browser that you and the agent share. Meta aims it at errands and goals: shopping, bills, travel, calendars.

OpenAI dots are always-on agents inside ChatGPT, powered by GPT-6 Astra. Our own guide to what dots do covers the launch, including the fact that the dot.com domain was already redirecting to xAI's Grok Bot page when OpenAI announced them.

What Grok Bot, Muse and dots cost per month

This is the table the pricing searches are really asking for. Every figure comes from a vendor page read on 1 October 2026: xAI's Grok Bot page, xAI's plans page, Cursor's pricing, Meta's Muse subscription page and ChatGPT pricing.

The Muse rows come straight from Meta's own subscription page, read on 1 October 2026.

Three caveats sit under that table. xAI lists SuperGrok Heavy and Cursor's Pro+, Ultra and Premium tiers as eligible for Grok Bot, but the pages we read show no price for them, so none appears above. Past the included allowance, xAI says extra Grok Bot usage is billed at token cost.

The dots Business Premium figure comes from The Decoder, not from OpenAI's own page. OpenAI says extra dots and higher speed will be sold later.

How setup works on day one

Setup tells you how much work each agent hands back before it saves any.

Grok Bot starts with a desktop download and a Bot you create and name. You log it into your apps once, then message it like a teammate. xAI says there are no workflows to build first, and the agent can learn a routine by watching you do a task a single time.

Muse starts with a Meta account. You connect the apps you want, such as email and calendar, and choose what Muse can see and do. Meta says that if a task needs a tool that does not exist, Muse builds one.

A dot starts in the ChatGPT desktop app. It takes context from your ChatGPT memory, then uses Codex and your connected tools. You pick which apps it may reach and how you like work done, per OpenAI's dots page.

The common step is trust. Each agent wants real logins on day one, so the first hour is mostly deciding what to connect. Start with a read-only mailbox, not your bank.

What each plan really costs once you count what you already pay

Plan price is not the cost of the agent. The cost of the agent is the extra money you spend because of it.

If you already pay for Cursor Pro to code, Grok Bot adds nothing. The same is true for a SuperGrok subscriber, and xAI says Bot usage does not eat into your Grok or Cursor allowance. For anyone else, the entry is Cursor Pro at $20 a month, since no standalone plan exists.

Muse starts at zero and tops out at $100. The top plan costs 5 times Power and buys 6 times the weekly tokens, so the heaviest tier is the better rate per token. Meta's help page says the paid tiers raise the weekly token allowance, so a heavy user buys headroom rather than new features. That makes the ladder easy to reason about: pay only when you hit the limit.

A dot adds nothing if you are already on Pro or Business Premium. Otherwise it means moving to a plan whose price we could not read. A Business Premium seat is reported at $100 to $125, the dearest single-seat route of the three. Set that against the agents sold on their own terms, such as Manus and ChatGPT Work.

What each agent can touch, and when it has to ask

An agent that signs into your real accounts can do real damage. Approvals are therefore the part of the spec worth reading closely.

Grok Bot is the loosest on paper. xAI says every Bot on your account shares one persistent computer, so files, browser and logins are shared and isolation is per user, not per Bot. Bots act under your own logins, which means a mistake inherits your access. xAI adds that sensitive actions can go through Auto Review before they run.

Muse is the most specific about money. Meta says Muse checks with you before sensitive actions such as sending an email or making a purchase. It says Muse pays through Link, which generates a one-time card so your real card stays hidden, and that Muse cannot see passwords stored in its credential vault. A separate Sentinel agent has to approve anything that reaches the internet.

Dots give you the most dials. OpenAI says you choose which apps to connect, decide whether to connect your own computer and set Custom Rules that allow, require approval for or block specific actions. Actions that could affect your accounts go through Auto-review, and steps like changing your password stay with you.

Where each one reaches you

Where an agent lives decides whether you actually use it.

Muse is the only one built around a messaging app people already open forty times a day. Meta says you can chat with it directly in WhatsApp, and that is a real distribution advantage over an app you must remember to launch.

Grok Bot asks you to install a desktop client. xAI's download button points at macOS, and the FAQ lists Windows and the iOS app too. There is no WhatsApp or Slack route on the pages we read.

Dots sit inside ChatGPT, and OpenAI lists Slack and Teams too. Per OpenAI's launch details, one dot reaches more than 4,000 apps through plugins. For a company that already runs on those chat tools, that is the shortest path from a request to a result. Details live on our dots page.

Where each one falls short

Every agent here has a weak spot you can name from the vendor's own pages.

Grok Bot has no free tier and no standalone plan, so you need an eligible subscription first. Its shared computer also means one place holds the files, browser and logins for every Bot you run. If that account is compromised, every Bot is exposed together.

Muse has no search across your own notes and documents, so it helps least when your work is scattered writing rather than bookable errands. It is also limited by region. It launched in the US only, and the small-business post later mentioned Canada.

Dots are the hardest to get. Access is limited to Pro and Business Premium in eligible markets, and Enterprise, Edu and Healthcare workspaces join through a beta that an admin must switch on. Texting a dot is listed as coming soon. A free ChatGPT user cannot try one at all.

The model underneath each agent

The model sets a ceiling on the work. The vendors name them, and that is where our sourced facts end.

Grok Bot runs on Grok 4.6, listed on xAI's own pricing page. OpenAI says dots are powered by GPT-6 Astra. Muse runs on Muse Spark, Meta's own model.

We have no agent-completion benchmark for any of the three, so we do not rank the models by how well they run these jobs. The model leaderboard shows what the models score on published tests. Those tests do not measure a week of unsupervised errands. Read model names as a clue about who controls the stack and who sells the plan, nothing more.

Where Grok Bot wins

Grok Bot wins on price if you are already a customer, and on team use. It is the only one of the three that describes several Bots working in a group chat, passing work between themselves while you judge the result.

It also handles the tools that have no API. xAI says a Bot signs in and works across apps and websites with no clean API or MCP route, which matters to an operations lead with a legacy support desk. The routine feature, where a Bot watches you once and repeats the job, is the clearest saving of setup time among the three.

Where Muse wins

Muse wins on cost of entry and on everyday life. It also gets the plainest setup of the three. Meta is also the only vendor here with a free tier of its own, and its ladder is the plainest to read: a weekly token figure sits on both paid steps.

Meta also leads on the purchase flow. Link's wallet and its purchase protections for eligible orders are a first for an AI agent, per Meta. If your jobs are bills, bookings, groceries and reminders, Muse is built for exactly that. For a business owner, Meta added a bundle of small-business connectors on 29 September, covering more than a dozen tools.

Where dots win

Dots win inside the workplace. They are also the only one of the three that keeps a separate computer for each dot. The launch examples are engineering and sales jobs: watching customer feedback and returning tested changes, checking an RFP against product docs, reworking an investor deck. A dot can also hand coding work to OpenAI Codex or ChatGPT Work, which keep their own usage limits.

The permission model is the richest, and the Slack and Teams reach suits a team that wants one shared channel to the agent. OpenAI also previewed specialist dots with their own credentials for enterprise pilots.

What happens to your data

All three run your accounts on the vendor's computer, so the data question is real.

Grok Bot uses Cursor's SSO, auth and privacy mode, according to xAI. The cloud computer is encrypted in transit and at rest, there is a training opt-out and enterprise admins can set DLP and network controls. xAI says this on its FAQ.

Meta says Muse conversations and VM data are not shared with Meta's ad systems, and that you can opt out of training use. Meta also promises a Muse Confidential VM later this year, with a key only you hold. It has not shipped.

OpenAI says dots build on ChatGPT's protections and run on its most aligned model. The page we read does not state a training policy for dots, so check your plan's data controls before connecting a work inbox. Our autonomy test of agents with real money shows why that deserves care.

What it costs to switch

Switching is cheaper than it looks, because no agent here holds your core data. Your mail, files and calendar stay in the apps you connected. What you lose is the learned context: Grok Bot's saved routines, Muse's memory of your preferences and a dot's feedback history.

We found no export tool for that learning on any vendor page we read. Plan for a week of re-teaching if you move. The shared computer matters too: Grok Bot and Muse each keep one persistent machine per user, so anything left on it stays there until you delete it.

The strongest case against this verdict

The best objection is that fit-by-existing-subscription is a lazy rule. It tells you what is cheapest, not what works best, and an agent that finishes 70 percent of jobs at no extra cost can still waste more of your time than one that finishes 95 percent at $100.

That is fair, and we cannot settle it. The honest answer is to treat the free or included entry as a trial. Give each agent the same two small jobs for a week, count how many finish without rescue, and only then pay for headroom. If you are unsure which jobs to give it, Smart Match can narrow the shortlist from your stack and your tolerance for risk.

A fair trial takes a week and four rules:

1. Give every agent the same two small jobs, such as sorting a week of email and booking one reservation.
2. Keep approvals on for anything that sends, buys or deletes.
3. Count the jobs that finished without a rescue, and note the minutes you spent correcting each one.
4. Pay for more headroom only if the free or included allowance ran out before the week did.

What we could not verify

A few gaps are worth naming.

• The ChatGPT Pro price is unreadable on OpenAI's page, so none appears here.
• No vendor publishes how often its agent finishes a job unaided, and we found no independent test.
• The $125 Business Premium figure is a press report, not an OpenAI page.
• xAI lists SuperGrok Heavy and higher Cursor tiers as eligible, but the pages we read show no price.
• Meta's launch post says US; its 29 September post says US and Canada.

What would change this verdict

Three things would move it. A published price for ChatGPT Pro with a dot would re-order the cost picture. An independent completion test would let us rank the agents on results. And Muse's arrival outside North America would change who it fits.

Self-hosting is the other route to watch. Self-hosted options such as OpenClaw and Hermes Agent trade the managed computer for control, and our OpenClaw playbook shows the setup cost. For the wider field, see the best AI agents of 2026 and the best agentic AI, or read what Manus is for. Meta's wider agent bet is in our strategy piece.

The next thing to watch is the week OpenAI's extended first-month dot limits end, because that is when the real monthly price of a dot shows up.

Frequently asked questions

How much does Grok Bot cost?

Grok Bot is not sold alone. It is included with SuperGrok at $30 a month, SuperGrok Plus at $100, Cursor Pro at $20 and Cursor Teams Standard at $40 per user. It has its own usage allowance, and xAI says extra use is billed at token cost.

How much does Meta Muse cost?

Muse is free with a usage limit. Meta's help page lists Power at $20 a month for 500M tokens a week and Maximum at $100 a month for 3B tokens a week. Paid plans raise the weekly allowance.

Do OpenAI dots cost extra?

OpenAI includes a first dot in ChatGPT Pro and Business Premium. Business Premium seats are reported at $125 a month, or $100 on annual billing. The Pro price is unreadable on OpenAI's page, and OpenAI says extra dots will be sold later.

Which always-on agent is safest?

No source we found ranks them on safety. Each vendor describes approvals before sensitive steps, and Muse and dots document the most detail. Check each plan's data controls and start with low-risk jobs.

Can I use these agents in WhatsApp or Slack?

Muse can be messaged directly in WhatsApp, per Meta. Dots are listed for ChatGPT, Slack and Teams. Grok Bot is offered through a desktop app and iOS, with no chat-app route on the pages we read.

Covered in this guide

• Grok Bot: xAI's always-on AI agents that sign into your existing apps and finish multi-step jobs, bundled with eligible SuperGrok and Cursor plans.
• Muse: Meta's personal AI agent books travel, negotiates bills and fills out forms from your accounts; launched Sept 8, 2026 with free, $20 and $100 tiers.
• OpenAI dots: OpenAI's always-on agent in ChatGPT runs on GPT-6 Astra with its own cloud computer, and asks approval before consequential actions.
• ChatGPT: ChatGPT is OpenAI's AI assistant, with 1.2 billion weekly users by OpenAI's count, the GPT-6 model family and plans that run from a free tier to a premium Pro tier.
• ChatGPT Work: OpenAI's ChatGPT Work agent plans and finishes multi-step projects across connected apps and files, launched July 9, 2026 for Business and Enterprise plans.
• GPT-6 Astra: OpenAI's flagship model, launched September 2026 as the first ever rated at the Preparedness Framework's Critical cybersecurity level.
• Grok: Truth-seeking AI chatbot with real-time web and X integration powered by advanced reasoning.
• Grok 4.6: Grok 4.6 (Aug 2026) is xAI's 500K-context reasoning model, matching the top score on the Artificial Analysis Intelligence Index.
• Hermes Agent: Hermes Agent is Nous Research's MIT-licensed self-improving AI agent, past 212,000 GitHub stars by July 2026. Self-hosted, model-agnostic, and free to use.
• Manus: Autonomous AI agent that plans and executes multi-step work from one prompt: research, documents, apps and video. The September 2026 release adds Cascade, Automations and Studio. Free tier available.
• Muse Spark: Muse Spark 1.3, released September 2, 2026 by Meta AI Research, is an agentic model built for long coding, long documents, and tool-heavy work.
• OpenAI Codex: Clones a GitHub repo into a sandbox, writes and tests code, then opens a pull request; OpenAI reported 20 million active users in August 2026.
• OpenClaw: OpenClaw is a local-first, open-source AI agent scoring 48.7% on SWE-bench Verified. Self-hosted and free, with a paid cloud tier available.
• xAI: Elon Musk's AI company (roughly 4,000 to 4,900 employees) builds the Grok models and merged into SpaceX, taking the combined business public on Nasdaq as the largest IPO on record.

Sources