October 22, 2026 | AI | By Mahesh

Anthropic opened a research preview of Claude Duet on October 21, 2026, a standalone app on iOS, Android and the web where Claude works as a proactive agent for professionals and small business owners [1][2][4]. Duet is built on Claude Opus 5.5, sends a morning brief, checks connected accounts every few hours and asks for approval before it sends messages, pays for anything, signs documents or deletes data [2][3]. Access is by waitlist for US Pro and Max subscribers, with no separate price [1][4].

What Claude Duet is and what is new

Claude Duet reverses the usual chat pattern. Instead of waiting for a prompt, it surfaces what needs attention, completes tasks you assign and keeps working when you step away [1][2]. Each morning it summarizes the inbox and calendar and drafts next steps, and one ongoing text or voice thread handles both quick requests and projects that run for weeks [1][2]. Example jobs include scheduling meetings, building dashboards, finding unpaid invoices, preparing payroll for review and getting contracts ready to send [1][6].

Memory, skills and connected apps carry over from an existing Claude account. 9to5Mac describes a Tasks tab for tracking work and a Make tab for finished files, and says only the user can mark a task as finished [3]. During the preview Duet is separate from the main Claude app, and Anthropic says early feedback will shape how the two work together [2].

"A proactive agent that understands what needs your attention."

Anthropic, Claude Duet product page, as quoted by AI Weekly [4]

Under the hood: model, cloud computer and approvals

Duet runs on Claude Opus 5.5 and does its work on a virtual machine in the cloud with its own browser, which lets it keep going when the user's devices are off and lets it use sites that have no connector [2][3]. Approvals, logins and safety checks run on that cloud computer, which Anthropic says keeps sensitive details private, and users can see what Duet has done and steer it [1][6]. TestingCatalog notes that users cannot yet watch the browser live [2].

Purchases go through Stripe Link or Shopify Shop Pay with approval steps, and Link uses virtual-card handling. Securities trading and transfers between accounts are prohibited [2]. Credential handling follows the design of Claude Managed Agents, action controls use the classifier from Claude and Claude Code, and a separate classifier blocks sharing of credit card numbers and government IDs. Health and personal finance data needs explicit instructions, and Anthropic says red teams, an outside security firm and a bug bounty program tested the system [2].

Evidence: what has and has not been shown

Anthropic has published no accuracy, failure-rate or task-completion figures for Duet, and Tech Insider notes that neither Anthropic nor Meta has released independent failure-rate data for their agents [4][5]. TestingCatalog and CellCog both describe features from Anthropic's pages and not hands-on testing, since access is waitlisted [2][6]. Details here therefore reflect Anthropic's claims.

Tech Insider compares Duet with Meta's Muse. Duet prepares work and waits for approval, while Meta describes Muse as moving toward a goal more independently, including browser actions and form filling. The article frames the trade as speed and reach against safety by design, and adds that approval gating only helps if users actually review what they approve [5]. Depth Grid covered Meta's model in Meta Muse Wonder Launch: 1M Context, $4 API Price, Benchmarks.

Pricing, availability and how to access it

Duet is included in Claude Pro and Max and has no separate price. It shares the plan's usage limits, background checks count toward them and larger projects need more model turns, so 9to5Mac reports that Anthropic warns users may reach limits sooner than with chat alone [1][2][3]. Access is granted in waves through a waitlist for US subscribers, on iOS, Android and claude.ai. Team and Enterprise plans are not included yet, and Anthropic plans to expand to them [3][4].

Anthropic has not said whether pricing will change when the preview ends or when Duet will reach other countries [5][6]. The product page lists connectors such as Gmail, Google Calendar, Slack, Stripe, QuickBooks and HubSpot, and testing may reveal gaps in the long tail of the 6,000 listed integrations [1].

Limitations, trade-offs and alternatives

Anthropic says Duet may misread a thread, miscalibrate what is urgent or miss something that arrives between check-ins. Personal errands are less developed than business tasks, and users cannot yet view the browser [1][2]. Each background check uses plan limits, and a proactive agent with access to email, calendar and payments widens the risk if approvals are skimmed.

The closest named alternative is Meta's Muse, and 9to5Mac's headline also names ChatGPT Dots, which its article does not examine. Depth Grid has not tested any of them [3][5]. Readers who track Anthropic's wider push can also see Claude Haiku 5.5 Launch: $0.10 Input, 1M Context, Benchmarks and the White House AI incident reporting push. Teams on Pro or Max should wait for hands-on reports before routing invoices or payroll through it.

Sources

1. Anthropic. "Claude Duet," product page. claude.com
2. TestingCatalog. "Anthropic opens Claude Duet preview for US Pro and Max users." testingcatalog.com
3. 9to5Mac. "Anthropic takes on ChatGPT Dots and Meta Muse with Claude Duet," October 21, 2026. 9to5mac.com
4. AI Weekly. "Anthropic Opens Claude Duet, an Opus 5.5 Agent, to US Pro and Max," October 21, 2026. aiweekly.co
5. Tech Insider. "Claude Duet: Anthropic's Opus 5.5 Agent Rivals Muse." tech-insider.org
6. CellCog. "Claude Duet: Anthropic's Proactive Agent, Explained" (from a competing AI agent vendor). cellcog.ai

Read More on Depth Grid

• Meta Muse Wonder Launch: 1M Context, $4 API Price, Benchmarks
• Claude Haiku 5.5 Launch: $0.10 Input, 1M Context, Benchmarks
• White House AI Incident Reporting and Anthropic
• Mistral Large 4 Preview: 1.05T Open-Weight MoE, Specs and Pricing

Article by Mahesh | Depth Grid