Anthropic launches Claude Duet, an always-on agent, in research preview for US Pro and Max users

Anthropic on October 21, 2026 announced Claude Duet, an always-on AI agent that manages tasks across connected apps and keeps working after the user closes the app. It is a research preview for Pro and Max subscribers in the US, with access granted in waves through a waitlist.

Availability and cost

• Eligible plans: Pro and Max, US only. Anthropic says Claude Duet is not available on Team or Enterprise plans yet, and that it plans to extend access to them.
• Price: No additional cost. Usage, including background work, counts toward existing plan limits. Anthropic warns users may hit those limits "sooner than with chat alone."
• Platforms: Standalone iOS and Android apps, plus the web at claude.ai, for invited users.
• Underlying model: Anthropic says Claude Duet is powered by Claude Opus 5.5. Context window, per-token pricing, benchmark scores and training cutoff for this deployment have not been disclosed.

How it works

Users message Claude Duet as they would a person, by text or voice. Anthropic says the main thread persists, and users can open separate channels for different topics that share the same memory. It uses the same memories, skills and connectors as regular Claude and can search past conversations.

Claude Duet runs on its own cloud-based computer, which is what allows it to continue working while the app is closed. For sites without a connector, it uses its own cloud browser.

Anthropic describes these workflows:

• Morning brief: Each morning it reviews email and calendar, picks what most needs attention, and can start first steps such as drafting replies for review.
• Follow-ups: It sends updates during the day and follows up on threads that have gone quiet.
• Larger jobs: Users describe an outcome and Claude Duet decides which apps to use and how much effort is needed. Results come back as editable files, such as a research brief, financial model, slide deck or dashboard. Finished work appears in a Make tab.
• Plugins: It supports plugins, such as a small business plugin, and can suggest ones that fit the user's work.
• Tasks tab: According to the support page, a dedicated tab tracks ongoing work and flags tasks needing attention. Only users can mark tasks as finished.

Payments and safeguards

Claude Duet integrates with Stripe to check for unpaid invoices and draft payment reminders. It also works with Link and Shop Pay to make purchases with user approval. Anthropic says it is designed to ask before sending a message, making a payment, signing a document or deleting anything. This is Anthropic's design claim, and no independent testing has been reported.

What this means

Claude Duet is a product built on an existing model, not a new model release. The notable shift is architectural: a persistent agent with its own cloud compute, shared memory and background execution, rather than a chat session that ends when the user leaves. Anthropic is positioning it against OpenAI's ChatGPT Dots and Meta's Muse, per the source's framing.

The usage-limit caveat is the practical constraint. Background work draws from the same quota as chat, and Anthropic has not published how much a typical always-on workload consumes. Whether Pro and Max limits can sustain daily briefs, follow-ups and large file-generating jobs is the open question for the preview.

The confirmation-before-acting design matters most where Claude Duet touches payments, signatures and deletions. How reliably it holds that line across connected apps will determine whether it is trusted with those actions. The US-only, waitlist-gated rollout and the exclusion of Team and Enterprise plans suggest Anthropic is limiting exposure while it gathers real-world data.