Claude Duet: Anthropic Opus 5.5 Agent Rivals Muse

Anthropic opened a research preview of Claude Duet on Tuesday. This AI agent works with the company's Claude assistant and can handle tasks without needing frequent instructions from the user. The Verge first reported the news. The launch puts Anthropic in direct competition with Meta's Muse family of agents, which Meta has been expanding quickly since early September. Claude Duet became available on October 21, 2026, to Claude Pro and Max subscribers in the United States who join a waitlist. It runs on Anthropic's Opus 5.5 model.

This launch comes during a major competition in consumer AI. The question is no longer just which chatbot gives the best answers. It is which AI agent people can trust to do things for them while they are busy with something else. Anthropic, Meta, and OpenAI are all working toward this goal in different ways. Claude Duet is Anthropic's clearest answer so far to a question the industry has been exploring for more than a year.

What Claude Duet Actually Does

Claude Duet is built around a simple idea: people often need an assistant to finish tasks, not just have conversations. According to Anthropic's product page, cited by Unite.AI, Claude Duet finds things that need a user's attention, handles tasks of different sizes, keeps working while the user is away from their device, and asks for approval before making decisions that could have serious effects. This last feature is central to Anthropic's approach: the AI can work independently, but it stops and asks a person to approve important actions instead of simply carrying them out.

Anthropic says Claude Duet can prepare morning briefings by checking a user's connected apps and highlighting important information. It can watch email inboxes and calendars for items that need a response, arrange meetings, build dashboards, draft message replies, and prepare reports or slide presentations. It can also find unpaid invoices and draft payment reminders, prepare payroll information for a manager to review, and get contracts ready to send. Users do not need to open another app or start a new conversation to do these tasks. Claude Duet brings the memory, skills, and connected apps from the main Claude app into the same conversation.

Opus 5.5 Is Doing the Heavy Lifting

Claude Duet runs on Opus 5.5, which Anthropic has described as its top-level reasoning model this year. Opus 5.5 recently ranked first in an independent comparison of large language models against GPT-6.1 Sol and Gemini 4 Argon. Our earlier coverage of the LLM leaderboard standings explains these results in detail. Opus 5.5 also led Terminal-Bench 4.0, a test of how well AI models handle long sequences of tasks in a computer terminal. We covered this result when Opus 5.5 hit 65.15% against rival models on long-running terminal tasks. Choosing Opus 5.5 instead of a smaller, cheaper model suggests that Anthropic sees proactive AI work as a difficult reasoning problem, not just a scheduling feature added to a chatbot.

This choice could also affect costs, but Anthropic has not shared details publicly. Running a powerful reasoning model in the background to prepare morning briefings, monitor email, and complete tasks with many steps uses much more computing power than answering one chat message. Anthropic has not announced a separate price for Claude Duet beyond the existing Pro and Max subscription plans. It has also not confirmed whether the price will change after the research preview ends.

Who Can Actually Use It Right Now

Access is limited. Claude Duet is available only to Claude Pro and Max subscribers in the United States, and they must join a waitlist. Anthropic has not included Team or Enterprise customers in the preview. This is notable because many of the advertised features, such as tracking invoices, preparing payroll information, and drafting contracts, seem especially useful for small businesses, not just individuals managing personal email.

Users can access Claude Duet through separate iOS and Android apps or through the existing website at claude.ai. This follows Anthropic's wider approach of making Claude available on different platforms instead of treating the agent as a completely separate product.

Why a Waitlist Instead of a Full Launch

Anthropic calls Claude Duet a "research preview" instead of releasing it to everyone. This follows the company's approach to other AI agent features. Giving an AI permission to work across someone's calendar, inbox, and financial documents creates risks that a chatbot answering questions does not face. The AI might misunderstand an email, put an invoice in the wrong category, or schedule a meeting incorrectly. These mistakes can cause real problems, not just produce an awkward answer. By starting with a limited group of paying subscribers who already use Anthropic's products, the company can watch for these problems and learn from them before giving more people access.

Meta's Muse: The Agent Anthropic Is Racing Against

Claude Duet is entering a market where Meta is already active. Over the past seven weeks, Meta has expanded Muse into what it now calls a frontier-class agent platform. Anthropic launched Claude Duet one day after Meta described Muse as having reached "the frontier." The timing makes the two announcements look closely connected.

From Muse Spark to Muse Wonder

Meta's push into AI agents gained momentum on September 2, 2026, when it released Muse Spark 1.3. The company said this version performed better on agent tasks and coding work. Six days later, on September 8, Meta introduced Muse, describing it as the "world's first personal AI agent built for everyone." Later that month, at Meta Connect 2026, the company announced plans to bring Muse beyond phones and computers to its smart glasses. This would let people use the agent throughout the day instead of only on a screen.

Meta took another major step on October 12, 2026, when it introduced Muse Wonder and Muse Wonder Full. According to Meta's own newsroom post, the company described them as its most capable models so far. They were built for personal and work tasks that require several steps, such as handling an insurance claim or managing a work project. Muse Wonder is available through the Meta Model API and Muse Code, while Muse Wonder Full is available only to Muse Code subscribers. CNBC reported that Meta considers Muse Wonder its most advanced option for powering digital assistants. This puts it in direct competition with Anthropic's Claude Duet.

Meta has described Muse as able to do more on its own than Anthropic has publicly said Claude Duet can do. According to a Meta statement reported by the Associated Press, "Once a person shares a goal with Muse, it helps them develop a personalized plan and coordinate their time and resources, then advances the work on its own. It can open a browser, fill out forms, and negotiate on their behalf." This description includes independent actions such as browsing websites, filling in forms, and negotiating for users. Anthropic has not yet described Claude Duet doing all these things. Its focus has been on monitoring information, drafting content, and preparing work for a person to approve.

Claude Duet vs Meta Muse Wonder: A Feature Comparison

Comparing the two products shows that they follow different approaches to proactive AI agents in late 2026.

The comparison shows a clear difference. Meta is building Muse to take more actions independently, while Anthropic is building Claude Duet to prepare work and wait for approval. Both approaches have risks, but the risks are different. An agent that negotiates or fills out forms for someone can save more time, but a wrong assumption could cause bigger problems before anyone notices. An agent that stops to ask for approval is designed to be safer, but that only helps if people read the requests instead of automatically clicking to approve them.

The Wider Proactive Agent Race

Anthropic and Meta are not the only companies competing in this area. The growing use of AI agents has already changed how people search for and use AI tools. Our earlier coverage of how OpenAI's Dots and Meta Muse are reshaping search behavior showed that people are starting to type fewer questions into Google and more instructions asking AI to complete tasks. We also compared prices and subscription plans in our report on ChatGPT's Dots, Meta Muse, and Claude's agent tier. Even before Claude Duet existed, the difference between free and paid access to proactive AI features was one of the biggest divides in the consumer AI market.

What makes this moment different is that all three major AI companies are now releasing versions of an agent that keeps working while the user is away. They are no longer treating this as a future goal. This is a major change from just six months ago, when most AI agent demonstrations focused on coding tasks. Anthropic's developer tools show the same change. Our guide to setting up the Claude Agent SDK running on Sonnet 5.5 explained how tools for carrying out tasks independently have spread across Anthropic's products for developers, before a consumer agent like Claude Duet arrived.

A Timeline of the 2026 Proactive Agent Push

Only seven weeks passed between Meta's first Muse release and Anthropic's response. That is a short time for a company that has usually moved more carefully than its two biggest competitors on consumer AI agents. The speed of this launch sends a clear message: Anthropic does not want Meta and OpenAI to define the market for proactive agents without a direct competitor.

Market Impact: What Investors and Enterprises Should Watch

For Anthropic, Claude Duet could help keep existing customers as well as attract new ones. Pro and Max subscribers already pay to use Claude. Giving them an assistant that starts tasks on its own, instead of only responding to requests, could give them a stronger reason to stay rather than switch to a competitor. Team and Enterprise subscribers are not included in the preview. This suggests Anthropic wants to study how individual users work with the feature before introducing it to businesses, which need more complex permissions and safety rules. Business accounts may involve shared calendars, shared inboxes, and financial systems used by several employees.

Meta has a different business model. Muse is part of a free, ad-supported platform with billions of existing users. Meta also offers Muse Wonder and Muse Wonder Full to developers through the Meta Model API and a separate Muse Code subscription. This gives Meta two possible ways to make money: consumer engagement data from people using free Muse features and subscriptions from developers who want to build with Muse Wonder. Fortune reported that Meta CEO Mark Zuckerberg told the Financial Times that the company had reached "the frontier" with Muse Wonder and Muse Wonder Full. He pointed to Meta's own published tests, which said the models performed better than competitors, including OpenAI's GPT-6 and Anthropic's models. A company's claims based on its own tests should be treated carefully until independent experts can check them. Still, Meta's claims show how important agent performance has become to its AI strategy.

Enterprise Safety Scrutiny Is Already Building

Anthropic's careful rollout also comes as people across the AI industry pay more attention to safety promises. Our earlier report on Anthropic's own proposed $1 billion AI safety threshold showed that the company has pushed for stronger safety standards across the industry while also releasing more AI features that can act independently. This tension is likely to remain an issue as Claude Duet moves from a US-only waitlist toward wider availability.

Historical Context: How We Got From Chatbots to Agents That Work Alone

It is worth remembering that this change has happened quickly. As recently as early 2025, the main type of AI product was still the simple chatbot: a user asked a question, the model answered, and the conversation ended. Coding tools were among the first AI systems to take on more work independently. They could write code, run it, and fix problems across several steps without asking a person to approve every line. According to NBC News, Meta's Chief AI Officer Alexandr Wang described the change this way: “It was only about six months ago that having a personal agent running on your own computer was the sign of someone who is completely AI-pilled and thinking nonstop about agents. But for that small group of nerds … it looked like the beginning of personal superintelligence. And today with Muse, we’ve brought that magic to everyone.”

The main difference between early coding agents and products like Muse and Claude Duet is the range of tasks they can handle. Coding agents work in a limited, text-based environment where success is often easy to measure. A personal assistant must work across email, calendars, financial documents, and messaging apps. Each tool can fail in different ways, and mistakes can have real consequences. This is why Anthropic's system, which asks for approval before important actions, and Meta's system, which works more independently toward goals, reflect genuinely different ideas about how AI agents should behave.

What People Who Study AI Agents Are Saying

Meta's leaders have spoken openly about where they think AI agents are heading. Mark Zuckerberg described his long-term vision for Muse this way, as reported by the Associated Press: “Everyone will have an exceptionally capable personal agent that understands you, your goals, and everything you care about. Your agent will work 24/7 on your behalf to improve your relationships, health, career, finances, home management, hobbies, and more.” This vision goes beyond what Anthropic has said about Claude Duet. Anthropic has described a tool focused on completing tasks, rather than an assistant that manages every part of a person's life.

Alexandr Wang has also explained how agents could help people get work done without watching their computers. He told CNN, “You can have it help you run your small business, or just get work done for you. You can walk away from your computer, and it keeps working for you on all the jobs you line up.” This is similar to what Anthropic says Claude Duet can do with invoices, payroll preparation, and contracts, although the two companies have different ideas about how much freedom the agent should have.

Wang has also emphasized safety when discussing what will make some AI agents more successful than others. He told NBC News, “My view is that trust and alignment are quickly becoming the most important capabilities that will differentiate agents and models. Any lab that doesn’t focus on alignment will fall behind.” His statement, as Meta's chief AI officer, closely matches Anthropic's decision to require approval for important actions, even though the two companies have designed their products differently.

Five Predictions for the Next Six Months

• Anthropic may open Claude Duet's waitlist to more countries after collecting enough information from the US research preview to improve how the system asks for approval.
• Team and Enterprise access may arrive within a few quarters rather than taking years. Anthropic has already described features for invoices, payroll, and contracts that could help small businesses.
• Meta will likely continue bringing Muse to more devices, building on its Connect 2026 announcement that the agent would come to smart glasses. This gives Meta a way to reach users that Anthropic and OpenAI currently lack.
• OpenAI is unlikely to stand still. Our earlier coverage of Dots versus Muse showed how quickly the two companies have worked to match each other's features. A similar push by OpenAI to expand its proactive agent features within the next two quarters would not be surprising.
• Independent testing of Meta's claims about Muse Wonder's leading performance will become an important issue. Meta's own comparisons with GPT-6 and Anthropic's models have not yet been checked by an independent third party.

The Trust Question Nobody Has Fully Answered

Both Anthropic and Meta are asking people to let AI handle important parts of their digital lives. These include inboxes, calendars, payroll information, and contracts. Muse can also use browsers and submit forms. Anthropic believes people will trust an agent more if it stops and asks before taking an important action. Meta believes people may trust an agent more if it completes tasks without interruptions, as long as it delivers the result they wanted. Neither company has published independent data showing how often its approach makes mistakes. The real test of Claude Duet and Muse Wonder will therefore not be their lists of features, but how well they handle millions of ordinary, complicated tasks over the coming months.

For now, Claude Duet is available only in the United States through a waitlist for its research preview. Muse Wonder is further along in its rollout to developers and consumers. This gives Meta a head start in actual use. It also means Meta's claims about reaching the "frontier" will be tested by a much larger number of users, and sooner, than Anthropic's claims about Claude Duet.

Frequently Asked Questions

What is Claude Duet?
Claude Duet is an AI agent from Anthropic that finds tasks needing attention, works on them independently, continues working while a user is away, and asks for approval before taking actions with important consequences. It is currently available as a research preview.

Who can access Claude Duet right now?
Only Claude Pro and Max subscribers in the United States can join the waitlist to get access. Team and Enterprise customers are not included in the first preview.

What AI model powers Claude Duet?
Claude Duet uses Opus 5.5, Anthropic's top-level reasoning model. This model has also led independent benchmark comparisons this year, including Terminal-Bench 4.0.

Does Claude Duet cost extra on top of a Claude subscription?
Anthropic has not announced a separate price for Claude Duet. Eligible Pro and Max subscribers can currently access it by joining the waitlist, and no additional fee has been confirmed.

How is Claude Duet different from Meta Muse?
Claude Duet is designed to prepare work and draft content, then ask the user for approval before taking important actions. Meta says Muse can work more independently toward a goal, including opening browsers, filling out forms, and negotiating on a user's behalf.

What platforms support Claude Duet?
Claude Duet is available through separate apps for iOS and Android and on the web at claude.ai.

When did Meta launch Muse Wonder?
Meta introduced Muse Wonder and Muse Wonder Full on October 12, 2026. The company described them as its most capable models so far for personal and professional tasks that require an AI agent.

Is Claude Duet available outside the United States?
Not yet. The research preview is limited to Claude Pro and Max subscribers in the United States who join the waitlist. Anthropic has not confirmed a date for international access.