What is Claude Duet?
Claude Duet is Anthropic's personal AI agent. It runs on Claude Opus 5.5, the model is not selectable, and it lives in a standalone app for iPhone and Android, plus the web at claude.ai. Anthropic opened it as a research preview on October 21, 2026, with a post titled Introducing Claude Duet. Coverage of the launch describes a waitlisted preview with invitations going out in waves, limited to the US, and limited to Claude Pro and Max subscribers.
Duet's job is to watch your connected apps, notice what needs attention, and act with your approval. It checks your apps every few hours, prepares a morning brief, and messages you only when something needs a decision. Anthropic positions it for professional and small business work first, not for general errands: think invoices, client follow ups, calendar conflicts, and business admin, rather than dinner reservations and gift shopping. An early explainer puts it plainly: Duet is aimed at the workday, and at the small business owner who runs it alone.
Muse is Meta's personal agent, and it takes the broader brief. It launched in the US on September 8, 2026, and covers shopping, calls, planning, messages, and briefings across your whole day. It has a free tier, it is live in five countries, and it needs no waitlist and no existing subscription. The two agents overlap on briefings, follow ups, and business admin, but they start from different places: Duet from professional work, Muse from everyday life.
A note on names, because the Claude family is easy to confuse. Claude Duet is the personal agent. Claude Code is the coding tool that runs on your own computer. Claude Pro and Claude Max are the subscription plans. Duet runs on Opus 5.5 and you cannot pick a different model for it. If you have read our Muse vs Claude comparison, that guide compares Muse with the Claude chatbot. This guide compares Muse with the Duet agent, which is a different product with a different job.
How is a research preview different from a shipped product?
The word preview does real work here, and it is worth understanding before you compare features. A research preview means Anthropic is still learning how people use the product, and the product can change quickly, break, or lose features without warning. Invitations go out in waves, so two subscribers on the same plan can have different access on the same day. Support exists, but the product is explicitly unfinished, and Anthropic published a safety and privacy page for Duet on October 23, 2026, two days after the preview opened, alongside a support article called Get started with Claude Duet on October 22.
Muse is a shipped product. It launched publicly, it has public pricing, it is in the app stores, and anyone in a launch country can download it and start today. Shipped does not mean finished: Muse is also young, and it changes often. But the relationship is different. With Muse, you are a user of a public product. With Duet, you are a participant in a research preview, and your use helps shape what the product becomes.
That difference shows up in the details. Duet admits come with one full limit reset, covering both the five hour and the weekly limits, available until November 30, 2026, according to launch coverage. That is a preview gesture: a fresh allowance while Anthropic watches how the product is used. Muse has no equivalent because it needs none: the free tier and the paid plans are the standing offer, not a preview allowance. If you need something stable this week, a shipped product is the safer bet. If you want to shape a new agent and you already pay for Claude, the preview is the interesting seat.
The preview label also affects how much you should trust any comparison, including this one. Duet gained Claude Code sessions in the app after its first week, which changed what it could do. More changes will follow. Treat every Duet claim in this guide as dated November 13, 2026, and check Anthropic's own pages before you act on it. The same caution applies to Muse, but Muse's public launch and public pricing make its facts easier to pin down.
How does Duet work day to day?
Duet's day starts with a morning brief. It checks your connected apps every few hours, pulls together what changed, and prepares a summary of what needs your attention: the invoice that is overdue, the client who replied, the calendar clash next week. Then it stays quiet. Duet messages you only when something needs a decision, which is a deliberate design choice: an agent that pings you all day is a notification feed, not a helper. A walkthrough of Duet's daily rhythm describes the pattern as check, brief, wait, and act on approval.
When Duet acts, it asks first. It shows you what it will do and waits for your approval before it sends, pays, signs, or deletes. That approval step is the core of the product, and it is also the most common complaint about it: in the nine task test covered later in this guide, Duet asked for permission four times, against once for each of the other agents, and it was consistently the slowest. Anthropic's bet is that caution earns trust faster than speed earns applause. Whether that bet pays off depends on how much you value being asked.
Muse's day is broader and louder. It also prepares briefings: our guide to the Muse morning briefing shows a small business owner starting the day with orders, messages, and priorities pulled together. But Muse also shops, calls businesses, plans trips, and builds Spaces, the custom screens that track the things you care about. Muse asks before sensitive actions too, and you choose whether an approval lasts once or always. The difference is scope: Duet watches your work apps and waits. Muse runs your errands and reports back.
Both agents keep context across sessions, so a thread you start on Monday is still there on Friday. Both learn your preferences over time. The day to day difference is really about what you hand over. If you hand over business admin and want to be asked before anything moves, Duet's rhythm fits. If you hand over your whole day and want the agent to keep moving with light touch approvals, Muse's rhythm fits. Neither is hands off. Both need you to check the work that matters.
What does each one cost?
Duet costs nothing extra during the preview, but it is not free. You need a Claude Pro or Max subscription to be invited at all. Claude Pro is $20 a month on monthly billing, or $17 a month billed annually. Claude Max starts at $100 a month. Duet shares your plan's usage limits, so heavy Duet use eats the same allowance as your Claude chats. Team and Enterprise plans are excluded for now, so a business on a team plan cannot get Duet yet, however much it pays. Pricing coverage from launch week confirms the plan ladder and the exclusion.
What Duet costs after the preview is unpublished. Anthropic has said nothing about a Duet price, a Duet tier, or whether preview users keep access on the same terms. That is normal for a research preview, but it means you cannot plan a budget around Duet today. If the preview ends and Duet becomes a paid add on, the price could be anything. If you build a business workflow on Duet now, build it knowing the price is unknown.
Muse is free to start, with a weekly usage limit, and its paid plans buy more usage, not more features: Power at $20 a month, and Maximum at $100 a month. The free tier and the paid tiers get the same agent, the same connectors, and the same features. A new account that redeems an invite code within 48 hours of joining also gets 1 billion Muse tokens, shared with the person who gave the code. The pricing is public, the plans are public, and the free tier is a standing offer, not a preview allowance.
Side by side, the entry cost is the same number with a different meaning. Claude Pro at $20 a month buys you a chance at a Duet invitation, if you are in the US and your wave has arrived. Muse Power at $20 a month buys you more usage of a product you already have, in any of five countries, today. And Muse free at $0 buys you the whole agent with a weekly limit, which has no equivalent on the Duet side. If cost is the deciding factor, Muse wins on entry. If you already pay for Claude Pro or Max, Duet costs you nothing extra to try, once your invitation arrives.
Who can use each one?
Duet is US only, waitlisted, and limited to Claude Pro and Max subscribers. Three gates, all at once. You need to be in the US, you need a qualifying plan, and you need an invitation, which arrives in waves with no published schedule. Team and Enterprise subscribers are excluded for now, even in the US. If you are outside the US, on a free Claude plan, or on a team plan, Duet is closed to you today, and Anthropic has published no date for opening it wider.
Muse is live in five countries: the US since September 8, 2026, Canada since September 18, Japan since October 23, Singapore since October 28, and Mexico since November 5, 2026. Anyone in those countries can download Muse and start, free, with no waitlist and no existing subscription. Outside those five countries, Muse is closed, and Meta has published no country list or timeline. The gate is geography, not money and not a waitlist.
The overlap is the US, Canada, Japan, Singapore, and Mexico on the Muse side, and the US only on the Duet side. A reader in Toronto can use Muse today and cannot use Duet at all. A reader in New York on Claude Pro can use both, once the Duet invitation arrives. A reader in London can use neither: Duet is US only, and Muse has not launched in the UK. A reader on a Claude team plan in San Francisco can use Muse, but not Duet, because team plans are excluded from the preview.
If you are deciding today, availability may decide for you. Check the three Duet gates first: US, Pro or Max, invitation. If any gate is closed, the comparison is academic, and Muse is the agent you can actually open. If all three gates are open, the rest of this guide is the comparison that matters: what each one connects to, how each handles money, and which one fits your work.
What can each one connect to?
Duet connects to more than 6,000 apps through connectors and plugins, according to launch coverage citing TechCrunch. The named apps skew business: Gmail, Slack, QuickBooks, Stripe, Gusto, LinkedIn, Docusign, HubSpot, Canva, and Shopify, among others. The list tells you who Duet is for. QuickBooks and Gusto are small business finance and payroll. Stripe and Shopify are selling. Docusign is contracts. This is an agent wired for the back office of a small business, not for a social feed. A comparison published at launch names the same business skew.
Muse connects to email, calendar, messages, social apps, shopping, and more, through its own connector list. The skew is everyday life: Gmail and Google Calendar, Instagram and Facebook, WhatsApp, shopping and travel apps, and the tools a small business uses day to day. Our guide to Muse for small business finances shows how Muse handles invoices, expenses, and follow ups from connected apps. The two connector lists overlap on the big business apps, and diverge at the edges: Duet reaches deeper into business software, Muse reaches wider into life.
Connector count is a weak way to compare. Six thousand connectors sounds like more than Muse's list, but the number that matters is how many of the apps you actually use are connected, and how well each connection works. A connector that reads but cannot act, or that syncs once a day, is worth less than a connector that acts in real time. Before you choose, list the five apps your day runs on, and check each one against both connector lists on the vendors' own pages. The agent that covers your five wins, whatever the headline count says.
One more connection difference: Duet gained Claude Code sessions in the app after its first week, which lets it work with code on your own computer through the Claude Code tool. That makes Duet interesting for developers and technical founders who already run Claude Code. Muse has its own developer surface through Muse Code and the Meta Model API, but that is separate from the consumer app. For most readers, the code connection is not the deciding factor. For developers, it might be.
How does each one handle money and permissions?
Duet handles money through Link and Shop Pay, with a one time card issued for each approved purchase. You need a US payment method. Duet asks before it pays, and it shows you what it will do before it does it. Anthropic also draws three hard lines: Duet will not trade securities, will not move money between accounts, and will not use someone else's account, even with approval. Those refusals are built in, not optional, and they tell you how Anthropic thinks about the blast radius of an agent that can spend.
Muse handles money with the same shape: approval before spending, and a single use card for new merchants, so a store you have never bought from does not get your real card. Muse asks before it sends, buys, or books, and you choose whether each approval lasts once or always. The permission model is the part to learn carefully, because a broad approval granted once can keep acting long after you forget it. Our Muse safety guide covers the permission settings and the data controls in full.
The difference is emphasis. Duet is cautious by design: it asks often, it refuses whole categories of money movement, and it messages only when something needs attention. That caution costs speed, and it showed in testing, where Duet was the slowest agent and asked most often. Muse is freer by design: it moves faster, it covers more of your day, and it trusts the approval you give it. That freedom costs vigilance, and it showed in September, when a broad approval led to a public incident in which Muse shared a user's address with a stranger. Both designs are defensible. Neither is free.
For business use, Duet's hard lines may be the deciding factor. An agent that cannot move money between accounts and cannot trade securities is an agent with a smaller blast radius, which is what a careful business owner wants. For personal use, Muse's breadth may matter more: an agent that can shop, book, and call covers more of the day. Either way, the rule is the same: keep approvals narrow, check anything that touches money or other people, and never grant an always approval for a job you would not trust a stranger to do unsupervised.
What did the nine task test find?
The most useful head to head evidence so far is a real world test, not a benchmark. Claire Vo, host of the How I AI podcast, ran four personal agents through nine of her real tasks in an episode published on October 28, 2026: a Slack reply, a short video cut, a bug fix, a Figma thumbnail, inbox triage, a family newspaper, a homework explanation, shoe shopping, and a product launch video. The episode notes record the results task by task.
Duet finished last, with a score of 79 percent, behind Grok Bot at 93 percent, Muse second, and Dot third, according to a review that scored the episode. But last place hides a mixed record. Duet produced the best product launch video, matching the brand and value proposition and generating a voiceover. It found a second bug in the coding task and avoided duplicating work Dot had already done. And it failed outright on the short video cut, because it could not download the footage. It was consistently the slowest agent, and it asked for permission four times, against once for each of the others.
The caveats matter more than the ranking. The judging was not blind, and Vo said so on camera. Duet was in its first week, before Anthropic added Claude Code sessions to the app, so the Duet in the test is not the Duet of today. Duet could only see Vo's work calendar, while other agents had different access. Muse had no Slack or GitHub connected, so it completed seven of nine tasks, the same as Duet. And every agent needed human nudges. None of the four ran fully on its own.
The fair reading is that Duet is the careful agent: slower, more questioning, stronger on depth and judgment, weaker on speed and breadth. That matches its design. An agent built for professional work that asks before it acts will lose a speed test and win a judgment test, and that is exactly what happened. If your work rewards care over speed, Duet's last place finish should not scare you. If your day rewards speed and breadth, Muse's second place finish, with the fastest time on four of nine tasks, is the more relevant signal. For the same question against OpenAI's agent, read our Muse vs Dots comparison.
What are the limits and risks?
Duet's limits are the limits of a preview. It is US only, waitlisted, and closed to team and Enterprise plans. The price after the preview is unpublished. The model is fixed at Opus 5.5, with no choice. It is slow, and it asks often, which some users will read as care and others will read as friction. And it is young: the product you join today will change, possibly in ways that break a workflow you built on it. An explainer from launch week lists the same preview caveats, and adds that Duet's business focus means general errands are not its strength.
Muse's limits are the limits of a young shipped product. It is live in only five countries. The free tier has a weekly usage limit. Shopping has real gaps, including the Amazon block from September 2026. Broad permissions can go wrong if granted carelessly. And Muse has faced more public privacy scrutiny than Duet, including a dispute over Mac Messages access in September and a study finding it discloses 31 of 35 App Store data types. Our Muse safety incidents timeline covers each incident with Meta's response.
Both agents share the risk that comes with acting on your behalf: misunderstanding, stale information, and steps you would not have approved. Both need broad access to be useful, and that access is the risk as well as the product. The safe way to use either one is the same: start with low stakes jobs, check work before it reaches other people or money, keep approvals narrow, and read the vendor's own privacy and safety pages. Anthropic published a Duet safety and privacy page on October 23, 2026. Meta publishes Muse data controls inside the app, under Settings and Data controls.
One risk is specific to Duet and worth naming: preview lock in. If you build your business admin around Duet during the preview, and the preview ends with a price you cannot pay, or a change that breaks your workflow, you rebuild. That is the price of early access. Muse has no preview lock in, because the product and the price are public. If you are building something your business depends on, that difference alone may settle the choice.
Which one should you pick?
Pick Duet if you are in the US, you already pay for Claude Pro or Max, your work is professional or small business admin, and you value caution over speed. Duet fits the business owner who wants a morning brief, careful approvals, and hard lines around money movement, and who is happy to shape a preview product. Join the waitlist, and use the wait to list the five apps your business runs on, so you can check them against Duet's connectors the day your invitation arrives.
Pick Muse if you live in one of its five countries and you want an agent today, free, with no waitlist, covering work and life: briefings, shopping, calls, planning, and messages in one app. Muse fits the person whose day is bigger than their job, and the small business owner who wants a briefing and an errand runner in the same product. It is the only one of the two you can open right now without paying or waiting.
Pick both if you are in the US on Claude Pro or Max and you live in a Muse country. Use Duet for the careful business admin where its approvals and hard lines earn their keep, and Muse for the faster, broader work where speed and breadth matter. The nine task test ended in the same place for its host: different agents for different parts of life. There is no prize for loyalty to one agent. There is a prize for using the right one for the job in front of you.
And if you are outside the US, or on a Claude team plan, or on no Claude plan at all, the honest answer is that Duet is closed to you today, and Muse is the agent you can actually use, if you live in one of its five countries. If you live nowhere near either launch map, join both waitlists and decide in advance, so you are ready when the door opens. The comparison in this guide will still be the right frame: preview or shipped, cautious or broad, work only or whole day.
Get started
Muse is the agent in this comparison you can try today, free, with no waitlist. Open the app, connect the apps you use, and give it one small job. If you are new to Muse, redeem a free invite code within 48 hours of joining and both you and the person who shared it get 1 billion Muse tokens.
Get a free code from the community pool.
Frequently asked questions
- What is Claude Duet?
- Claude Duet is Anthropic's personal AI agent, running on Claude Opus 5.5 in a standalone app and on the web. It checks your connected apps every few hours, prepares a morning brief, and acts with your approval. It is aimed at professional and small business work first, not general errands.
- Is Claude Duet free?
- Duet costs nothing extra during the research preview, but you need a Claude Pro or Max subscription to be invited. Pro is $20 a month on monthly billing, or $17 a month billed annually, and Max starts at $100 a month. Duet shares your plan's usage limits, and the price after the preview is unpublished.
- Who can use Claude Duet?
- US only, Claude Pro and Max subscribers only, by waitlist invitation in waves. Team and Enterprise plans are excluded for now. If you are outside the US, on a free plan, or on a team plan, Duet is closed to you today, with no published date for opening wider.
- Is Muse free?
- Yes, with a weekly usage limit. Muse has a free tier, and its paid plans buy more usage, not more features: Power at $20 a month and Maximum at $100 a month. Muse is live in five countries with no waitlist, and a new account that redeems an invite code within 48 hours of joining gets 1 billion Muse tokens.
- Which is better, Muse or Claude Duet?
- It depends on the job and on access: Duet fits careful professional and small business admin for US Claude subscribers who value approvals and hard lines around money, while Muse fits a broader day, work and life together, and it is the only one of the two with a free tier and no waitlist. In Claire Vo's nine task test, Muse finished second and Duet finished last, with Duet in its first week and important caveats about access.
- Can I use Muse and Claude Duet at the same time?
- Yes, they are separate products from separate companies, and nothing stops you using both. A natural split is Duet for careful business admin and Muse for faster, broader work. Keep approvals narrow on both, and check anything that touches money or other people.
Built with Muse is an independent community project. Not affiliated with or endorsed by Meta.