Claude Duet vs OpenAI Dots vs Meta Muse

Compare Claude Duet, OpenAI Dots and Meta Muse from their launch material: computer, channels, work while away, approvals, access, cost and connections.

Claude Duet, OpenAI Dots and Meta Muse are personal agents that keep working after you stop typing. Each has its own cloud computer, reaches the apps you connect and comes back to you before a consequential step. Meta launched Muse on September 8, 2026, OpenAI introduced Dots on September 29, and Anthropic opened a research preview of Claude Duet on October 21. The real differences are in how each one works while you are away, where you talk to it, who can get it and how it reaches outside tools. This is a launch-day comparison from each vendor's own material, not a benchmark of task success.

The Claude Duet announcement, the OpenAI Dots announcement and the Meta Muse announcement support the rows above. Launch terms change, so check each vendor's current plan page before you buy on the strength of a launch article.

Where Claude Duet stands out

Duet's distinguishing idea is that it comes to you. It sends a morning brief from your email and calendar, checks your connected accounts at intervals, and messages you only when something needs a decision. It runs on Claude Opus 5.5 and decides its own effort, and it starts with everything you already set up in Claude: memory, skills and connectors are there from the first message, and it can search your past Claude chats. Larger work comes back as editable files in a Make tab, and a Tasks tab shows what is in flight.

Its approval model is the most explicit of the three. Duet is designed to stop before it sends a message, makes a payment, signs a document or deletes anything, and Anthropic names actions it refuses even with your approval: trading securities, moving money between accounts and accessing someone else's account. It also lists its own gaps plainly, including no Microsoft 365 yet, no texts or calls, and items that can arrive between check-ins. Our Claude Duet explainer covers all of it.

Where OpenAI Dots stand out

Dots live inside ChatGPT, so the same product family supplies the conversation, the plugins, Codex and ChatGPT Work tasks a dot can hand work to. Of the three, Dots have the clearest scheduler: you set recurring tasks and find them under Scheduled, and proactive research runs with read-only tools while you are away. They also reach you in more places, with Slack and Teams beside ChatGPT. Access is broader too. Dots are the only one of the three offered outside the US at launch, through Business Premium in every supported ChatGPT region. See OpenAI Dots explained for the detail.

Where Meta Muse stands out

Muse is the consumer agent of the three, reachable in its own app and in WhatsApp, which matters if WhatsApp is where your day already happens. Meta describes a dedicated Secure VM, a memory, ongoing goals and a system called Sentinel that checks every outside action, allowing it, denying it or sending it to you for approval. By Meta's published design, a credential you connect is stored in your own VM and inserted at the network boundary, so the agent handles a stand-in rather than the raw key. Most needs are free. Muse and Duet also share a payment rail: Stripe says Link powers payments for more than 200 personal agents, Muse among them, and it is the wallet built into Duet. Our Meta Muse guide covers how Muse reaches outside services.

How do they work while you are away?

This is the question that separates them most, and the answer decides which jobs each one should own.

Duet does not describe user-set scheduled tasks. It runs a morning brief and checks your accounts every few hours, "because checking nonstop would use up your plan's limits much faster," in Anthropic's words, and Anthropic warns that something urgent can arrive between checks. Dots expose a scheduler you control, with recurring tasks at the times you choose, and keep proactive research read-only. Muse describes background work and tasks on a schedule. Duet's background checks draw on its plan limits, while OpenAI kept dot usage out of plan allowances for the first month. And all three carry that work inside one agent's broad context, beside everything else it is doing for you.

That is the gap a separate AI worker is built for. A WorkerKit worker has one job, its own instruction, app grants, model and budget, and a schedule in its own time zone or a webhook that starts it. Each run leaves a receipt with its settled cost. The personal agent then reads the result rather than carrying the recurring job inside one broad context.

Which one should you try?

Start from where you already work. If Claude holds your memory, skills and connectors, and you want an agent that surfaces what needs you without being asked, try Duet, provided you are on a US Pro or Max plan and can wait for the waitlist. If ChatGPT is central and you want recurring tasks you schedule yourself, or you are outside the US, start with Dots. If your day runs in WhatsApp, look at Muse. Then give all three the same small job before deciding.

Compare one real task before choosing

A launch video shows what a product is designed to do. Your own workflow shows whether it does the job with the access and review you need. Pick a job that takes fifteen to thirty minutes and that a person can check, such as a briefing for tomorrow's meetings from recent email and a project document. Give each agent the same sources and the same definition of done.

For a standing version of the same job, a meeting prep worker or the executive daily briefing makes a useful baseline: a defined job, a schedule and a receipt to compare against each agent's output. The critical email finder tests the timing row directly, since it returns only the messages that cannot wait until tomorrow. Review each kit's permissions before installing it.

How each one reaches a WorkerKit fleet

The three take three different doors, and only two are documented end to end today.

For Duet, the Claude Duet guide adds both connectors in Claude and then asks Duet whether it can see them. Whatever that check finds, a worker can email its report to the inbox Duet already reads, and the setup guide walks through that route too. An earlier comparison, OpenAI Dots vs Meta Muse vs Grok Bot, covers Grok Bot as well.

FAQ

Which of the three is free?

None is free of limits. Muse covers most needs at no charge with subscriptions for more. Duet has no extra charge during its preview but needs a Pro or Max plan and draws on its limits, background checks included. The first dot is included in eligible Pro and Business Premium plans, and OpenAI excluded dot usage from plan allowances for the first month after launch.

Which one can I use outside the US?

At launch, Dots. OpenAI offers Business Premium across supported ChatGPT regions and Pro outside the EEA, Switzerland and the UK. Duet's preview requires a US Claude account, and Meta describes a US rollout for Muse.

Which one runs recurring work on a schedule I set?

Dots describe scheduled tasks you configure, and Muse describes tasks on a schedule. Duet describes a morning brief and check-ins at intervals rather than a schedule you set. For a job that must run at a set time with its own access and receipt, a separate worker schedule gives that job a boundary of its own.

Is Claude Duet available on Team or Enterprise plans?

Not yet. Anthropic says Team and Enterprise come later. Dots offer an Enterprise beta that a workspace admin enables, so for a business account today Dots are the one of the three with an organizational path.

Can all three share one WorkerKit fleet?

Yes, in principle, because the fleet is reached through MCP, the REST API and the dashboard rather than through any one assistant. Dots and Muse have documented routes. Duet's route runs through the connectors you set up in Claude, and whether those include custom connectors inside Duet is the one open question.

Sources: Anthropic's Claude Duet announcement, Get started with Claude Duet, Stripe on Claude Duet, OpenAI Dots, OpenAI's getting started guide for Dots and Meta Muse. All three are new, so check current terms before relying on any row.