Who Can Use Claude Duet and How Do Its Usage Limits Work?

Who can use Claude Duet, and how much of my usage limit does it consume?

Claude Duet is a research preview available to Claude Pro and Max subscribers in the US, and access runs through a waitlist. It does not have a separate allowance. Duet shares your plan's existing usage limits, and its background check-ins count against them. That means a Duet-heavy day can leave you with less room for regular chat, so you may hit your limit sooner than you would with chat alone.

What is Claude Duet?

Claude Duet is an Anthropic app where Claude acts as a proactive agent. Most chat tools wait for you to type something. Duet reverses that. It comes to you first.

According to Anthropic's announcement, it is powered by Claude Opus 5.5 and runs on its own computer in the cloud. It sends a brief every morning. Through the rest of the day it only messages you when something needs your attention. Anthropic frames the value of the product as its judgment about what needs you and what can wait.

That last point is the whole pitch. A proactive agent that pings you constantly is worse than no agent. The interesting question is whether Duet's filtering is good enough that you trust the silence.

Who gets access to Claude Duet?

Three conditions apply, based on what Anthropic has said so far:

If you are on a free plan, outside the US, or on a plan type not named above, the announcement does not offer you a path in. Anthropic has not said, in the material available, when or whether access will widen. Treat any timeline you see elsewhere as speculation unless it comes from Anthropic directly.

Being on the waitlist does not mean immediate access. It also does not guarantee a date. Expect staged invitations, which is typical for research previews, though the specifics of how Anthropic will admit people have not been spelled out.

Why is it called a research preview?

A research preview signals that the product is early. Anthropic is releasing it as a research preview to Pro and Max subscribers in the US through a waitlist. For a proactive agent, that makes sense. The hard parts are behavioral: when to interrupt, what to summarize, which tasks to handle quietly and which to escalate.

Those are judgment calls, and they are difficult to test without real inboxes, calendars, and routines. A limited rollout lets Anthropic watch how the agent performs before opening it more widely. It also means you should expect rough edges and changes to how it works.

How do Claude Duet usage limits work?

Duet does not come with its own bucket of usage. It shares the limits of the plan you already pay for. Anything Duet does counts against the same allowance you use for ordinary Claude conversations.

The detail that matters most is that background check-ins count too. With a normal chat assistant, usage tracks what you do. You send a message, you spend some allowance. With a proactive agent, the agent also acts when you are not looking. It checks in, evaluates whether something deserves your attention, and decides whether to message you. Even its background check-ins count toward your limit.

The practical consequence is simple. Because Duet shares your plan's limits, you may hit them sooner than with chat alone. If you rely on Claude for long coding sessions, document analysis, or heavy research, adding Duet to the same plan means those activities compete for the same pool.

Anthropic's announcement, as summarized in the source material, does not give a number for how much Duet consumes per day or per check-in. I would not trust any figure you see for that until Anthropic publishes one or you measure it yourself.

Is Duet's usage cost worth it?

That depends on what you use your plan for. A few things to weigh:

• Heavy chat or coding users have the most to lose. If you already run close to your limit, a background agent that draws from the same pool could push you over more often.
• Light users may have plenty of headroom, so the morning brief and occasional alerts may fit without trouble.
• Max subscribers presumably have more room than Pro subscribers, since Max is the higher tier, but the sources do not say how Duet scales across tiers.

A sensible approach during a research preview is to watch your usage during the first week. Note how quickly you reach your limit compared with your normal pattern. If Duet's brief and alerts save you real time, the trade may be fine. If they mostly produce noise, you are paying for it twice: once in attention and once in allowance.

How does Duet handle sensitive actions?

Duet asks before it acts on anything consequential. Before it sends an email or makes a payment, it waits for your approval. That is a sensible guardrail for an agent with its own computer and the ability to take actions on your behalf.

It also shapes the usage picture. An agent that stops for approval is doing less unsupervised work than one that executes freely, but it still has to do the thinking beforehand: reading context, drafting, and deciding what to surface. That work still counts.

How does Duet fit with other proactive agents?

Duet arrives in a crowded moment. The announcement coverage notes that Meta's Muse and OpenAI's dots have both launched since early September, and it mentions at least one other entrant. The shared idea is an assistant that initiates rather than waits.

What separates these products will likely be judgment quality and cost of use, not the concept. For Duet, the cost side is already partly visible: it shares plan limits and counts background activity. The judgment side is not yet known. Anthropic says the value lies in knowing what needs you and what can wait, and that claim needs real-world testing before anyone can call it proven.

What should you do while waiting for access?

If you qualify, join the waitlist. If you do not, there is little to act on yet. In either case, a few low-effort steps help:

1. Check which plan you are on and whether your account is set to a US region.
2. Note how close you typically get to your usage limit, so you have a baseline.
3. Decide in advance which tasks you want a proactive agent to watch, so you can judge its filtering against something concrete.