Claude Duet Is Anthropic's Answer to Dots and Muse. It Shares Your Pro Plan's Limits, Checks In Every Few Hours, and Is a US Waitlist.
Anthropic opened a research preview of Claude Duet on October 21: a proactive agent on Opus 5.5 with its own VM and browser, on iOS, Android, and claude.ai, included in Pro and Max for a US waitlist. The catches: it draws on the same usage limits as your chat, it checks in only every few hours because nonstop would burn the cap, it blocks trading securities and moving money even with your approval, and there is no Team, Enterprise, API, or separate price. We read Anthropic's article and the launch coverage.
By T.W. Ghost
Every lab now has an always-on agent. OpenAI shipped Dots in September, one per Pro plan with a cloud computer each. Meta's Muse runs on its own hardware. xAI's Grok Bot shares one computer across all your Bots. Google's Gemini agent is in private preview with no price. Today, October 21, 2026, Anthropic opened a research preview of its entry, Claude Duet, and it comes with a pricing model none of the others use: it does not have one.
Duet is a proactive agent on Claude Opus 5.5 with its own virtual machine and browser, running in a standalone iOS and Android app and on claude.ai. It writes morning briefs, runs background tasks, and asks for approval before it sends, pays, signs, or deletes anything. It is included in the Pro and Max plans for users on a US waitlist. And that word, included, is the whole story. We read Anthropic's article and the VentureBeat and CNBC coverage. Here is what held up.
What You Get, and What Meter It Runs On
| Claude Duet (research preview) | |
|---|---|
| Opened | October 21, 2026 |
| Model | Claude Opus 5.5 |
| Runs on | Its own VM and browser; iOS and Android app, plus claude.ai |
| Who | Pro and Max subscribers, US only, waitlist |
| Price | None separate; included in the plan, counted against the plan's usage limits |
| Cadence | Checks in "every few hours" |
| Approvals | Required before send, pay, sign, or delete |
| Hard blocks | Trading securities and moving money, even with your approval |
| Not available | Team, Enterprise, API, or a separate SKU; it cannot show you its browser yet |
Same meter as chat. Anthropic's own article says background work counts against your plan limits. Duet is not extra capacity bolted onto a subscription; it is a new way to spend the capacity you already have. That is the opposite of Dots, where dot conversations are exempt from ChatGPT limits, and it is why the next line exists.
"Every few hours" is a budget, not a feature. Anthropic says Duet checks in every few hours because running nonstop would burn through the cap. An agent that works for you in the background is throttled by the same quota that throttles your chats, which means a heavy Duet week is a light chat week.
The money rails are closed. Approval before paying is one thing. Anthropic goes further: trading securities and moving money stay blocked even when you approve. If you were hoping for a personal CFO, this is not it, and Anthropic says so in the launch article.
No enterprise door. VentureBeat reports no Team or Enterprise access, no API, and no separate price; Anthropic's page confirms the gaps by omission. For now Duet is a consumer product on consumer plans in one country.
How It Compares With the Others
| Claude Duet | OpenAI Dots | Grok Bot | Gemini agent | |
|---|---|---|---|---|
| Compute | Own VM and browser | Own cloud computer per dot | One shared computer across all Bots | Google Cloud, private preview |
| Included with | Pro and Max | Pro tiers and Business Premium | Premium tier and up | Gemini Enterprise, per press |
| Usage | Counts against plan limits | Dot conversations exempt from limits | Shared weekly pool, base undefined | No price published |
| Geography | US waitlist | Pro dots exclude EEA, UK, Switzerland | Global | Private preview |
| Money | Trading and transfers blocked even with approval | Approvals; extra dots unpriced | Payments shipped Oct 11 | n/a |
The comparison the launch invites is capability. The comparison that matters to a buyer is the meter: Dots give you separate capacity with a geo-fence, Grok Bot gives you a shared pool with no base number, Duet gives you your own existing cap.
Verified vs Unconfirmed: The Scorecard
| Claim | Verdict |
|---|---|
| Research preview Oct 21, 2026; Opus 5.5; own VM and browser; iOS, Android, claude.ai | Verified (Anthropic article) |
| Included in Pro and Max; US only; waitlist | Verified (Anthropic article; CNBC and VentureBeat) |
| Approval before send, pay, sign, delete; trading and money transfers blocked even with approval | Verified (Anthropic article) |
| Checks in every few hours; background work counts against plan limits | Verified (Anthropic article) |
| No Team, Enterprise, API, or separate price | Press (VentureBeat), consistent with Anthropic's page |
| Cannot show its browser yet | Verified (Anthropic article and VentureBeat) |
Who Should Use It
Join the waitlist if you are a US Pro or Max subscriber with recurring background work (morning briefs, inbox triage, research that runs while you sleep) and your chat usage leaves headroom in the plan. The approval model is conservative in the right places, and it costs nothing beyond the plan you have.
Skip it if your plan already runs near its cap, if you need it for a team or through an API, or if the job involves money moving anywhere; those doors are closed by design. And watch your meter the first week.
The frontier is elsewhere: on always-on agents the field is four products with four different answers to who pays for the compute. Duet's answer, you already did, is the clearest and the most limiting. Our model comparison has the current standings side by side.
Update, October 22, 2026
- October 22: SemiAnalysis reported on X what looked like an A/B test on two Claude Max 20x accounts: on the same day and model, one account had a 19% larger five-hour window (2,076M against 1,740M agentic tokens) but a 20% smaller weekly budget (5,266M against 6,579M), which SemiAnalysis maps to roughly $9,385 against $11,726 a month in its own API-equivalent arithmetic. We have not read the SemiAnalysis post itself; the numbers come from TechCityAuthority and CellCog quoting it, so treat every figure as secondhand. Earlier the same day, before that post, Anthropic's Boris Cherny and Lydia Hallie replied to users that limits had not changed; Anthropic has neither confirmed nor denied a test since. Sample size: two accounts. The reason it belongs here is the meter. Duet spends the same weekly budget, and if that budget varies by account, so does what Duet can do for you.
Sources
- Anthropic: Claude Duet
- VentureBeat: Anthropic's answer to Muse is here
- CNBC: Anthropic previews Claude Duet
- TechCityAuthority: Claude Max usage limits A/B test
- CellCog: Claude Max usage limits A/B test
*Not sure whether an always-on agent that spends your own plan is worth the headroom? Take the free 2-minute quiz and get matched. Then read about the agent that gets its own computer and its own blanks on the price list: OpenAI's Dots.*