# Claude Duet Features, Cost and Workflow Setup in 2026 By Carlos Mendoza · 2026-10-21 · Source: https://www.activepieces.com/blog/claude-duet-features-cost-and-workflow-setup-in-2026 --- [Anthropic](https://www.anthropic.com/claude/haiku) launched [Claude](https://claude.com/product/claude-duet) Duet **in 2026 as a research preview**. It's a proactive agent you message like a person, typed or spoken, that works in your connected apps, much as automation tools like [Activepieces](https://www.activepieces.com) work across theirs, and asks before sending messages, making payments, signing documents or deleting anything. ![Assistant holding an envelope up for approval](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/19b87269-8071-4da7-b264-9797ba9b2440/claude-duet-features-cost-and-workflow-setup-in-4a61d589.webp) Expect the behavior to shift, because Anthropic's own [support article](https://support.claude.com/en/articles/17631495-get-started-with-claude-duet) calls it a research preview. The table shows what each page confirms. Treat the first three as the maker's word and the fourth as a separate project. ## Claude Duet's 2026 agent launch | Source | What it confirms | Link | Checked | |---|---|---|---| | support.claude.com article | Description: message it like a person; it asks before sending, paying, signing or deleting | [Get started](https://support.claude.com/en/articles/17631495-get-started-with-claude-duet) | 2026-10-21 | | claude.com product page | Description: a proactive agent that works while you're away and waits for your ok | [Product page](https://claude.com/product/claude-duet) | 2026-10-21 | | claude.com resources article | Announcement: a research preview of a new app | [Announcement](https://claude.com/resources/articles/claude-duet) | 2026-10-21 | | GitHub repo | A different tool: a CLI for sharing a Claude Code session, by Eliran Geffen | [Repo](https://github.com/elirang/claude-duet) | 2026-10-21 | _Prices and plan limits checked against [support.claude.com](https://support.claude.com/en/articles/17631495-get-started-with-claude-duet) and [claude.com](https://claude.com/product/claude-duet) and [claude.com](https://claude.com/resources/articles/claude-duet) and [github.com](https://github.com/elirang/claude-duet) and [activepieces.com](https://www.activepieces.com/pieces/claude) and [docs.claude.com](https://docs.claude.com/en/docs/about-claude/models/overview) and [claude.com](https://claude.com/pricing) and [openai.com](https://openai.com/chatgpt/pricing) and [gemini.google](https://gemini.google/subscriptions) on October 21, 2026._ Don't mix up the last row with the app. Install instructions for a CLI are not install steps for Duet. ### How to get it, per the maker You can test it before anyone approves spend, since the plans that include Duet start at [$20 per month on Pro](https://claude.com/pricing). Max 5x is $100 per month, five times Pro. ![What Claude individual plans that include Duet cost](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/ded43591-afab-4658-bbd8-3903856e1deb/claude-duet-features-cost-and-workflow-setup-in-6a2ac01c.svg "Source: Claude") Max 20x is $200 per month, ten times Pro, so justify it with measured usage. If you later route approvals through [Activepieces](https://www.activepieces.com), its Cloud Free plan is $0 with unlimited flows and 1,000 credits a month, so that pilot can start on Free too. ## Why Claude Duet matters to teams automating work For automators, Duet matters for one reason. The maker's pages describe an assistant that **acts inside the apps your team already uses**, with a human sign-off at the risky moments, rather than one that only answers questions. That's also why a pilot comes first: the change lands on daily habits before it lands on any system. ### Where it touches an automated workflow Read the four confirmed behaviors as places Duet could replace a manual hand-off, not as places it replaces a workflow: - You message it by typing or by voice. - It works inside connected apps. - It **asks before sending, paying, signing or deleting**. - It reuses your account's memory, skills and connectors. Take the third item to your team. It means the person who triages the queue stays in the loop on the irreversible steps, so the habit you're teaching is "approve or decline", not "watch everything". ![Person choosing between two buttons for one flagged parcel](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/6686012e-5f0f-4fcc-a732-a50b94e1246d/claude-duet-features-cost-and-workflow-setup-in-a9f106fd.webp) The fourth means setup you've already done carries over instead of being rebuilt, so the first week is less about configuration and more about who owns the approvals. What Duet does is on the pages; how well it does it is not. Treat these as open until the maker answers them: - How reliably it finishes a multi-step job without stalling. - Whether approval prompts leave a record you can audit. - Whether an administrator can limit which connectors it may use. - What happens when a connected app changes underneath it. **Until the maker answers them, keep Duet away from work where a silent failure costs you.** ## What is measurably new in Claude Duet Three things are stated by Claude Duet's maker: **it runs on Opus 5.5**, it ships with a large catalog of connectors and plugins, and it works on its own computer in the cloud. We checked each against the maker's pages as of 2026-10-21 and add no benchmarks of our own. ### Capabilities the maker states Underneath, it is Opus 5.5. Anthropic positions it for long-running agentic coding and knowledge work, which is the kind of job an automation owner hands off and walks away from. That's also where the previous section's warning applies, because a long unattended run is where a silent failure hides. Your first pilot probably won't need custom integration code, because the connector and plugin catalog is large. Its size says nothing about how well any single connector behaves, so test the two or three your team touches daily before you trust the rest. Duet works on its own computer in the cloud, using the apps you connect. It can keep working after you close the app, and it doesn't require your laptop or phone to be on. Your IT contact should confirm early that your network permits that traffic, since **that's the unglamorous thing that stalls a rollout**. ### Claude Duet limits, waitlist and requirements Today, access is a waitlist, open to Pro and Max subscribers in the US. If your team sits outside the US or on a different plan, you can't pilot it yet. If you qualify, join the waitlist now and plan the first trial around whoever gets in, not around the whole team. If Duet fails partway through a job, the maker's pages don't yet say how it tells you. Until they do, treat that as unknown. ## Activepieces and Claude Duet: what we can confirm We couldn't confirm a Duet-specific release from Activepieces: no dated announcement and no dedicated integration we can link, only its general Claude integration. ### What shipped, when, and where to find it In [Activepieces](https://www.activepieces.com/pieces/claude), an automation builder, integrations are called pieces. Its catalog lists an [Anthropic Claude piece](https://www.activepieces.com/pieces/claude) (@activepieces/piece-claude) with 3 actions and 0 triggers. Zero triggers means Claude can never start a flow, so every workflow has to begin with something else, such as a schedule or a webhook. We found no date tying that integration to Duet, so treat any claim that it covers Duet as unverified until the page says so. Activepieces also serves each connected integration action as a tool on a per-project MCP server, so Claude, ChatGPT, Cursor or an agent you wrote yourself can call the very action a flow would run. MoneyGram, Moneypenny, Alan and FundingSocieties run it. ### Using it in a workflow today Since we haven't run Duet ourselves, this is the cautious route. Build the flow around the integration you can see, and check at the last step whether Duet is selectable. If it isn't, you've lost an afternoon of wiring at most, not a deployment. In the builder, an Every Hour Schedule trigger is selected, with the spreadsheet integrations open for the step after it. That's the shape to copy: the schedule supplies the trigger Claude lacks, and a sheet row gives each result a place a human can review. ![Activepieces flow builder showing a piece selector modal with spreadsheet integration options and a Schedule trigger step.](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/4be981c4-ec0d-4fde-af5f-4f549ed04004/how-webhook-triggers-detect-and-send-real-time-d-c908c50f.webp) The successful Generate Sample Data result in the settings panel is your proof the skeleton runs before any model is involved. Build that skeleton first and test it. Then add the Claude action and look for Duet in its options. Until Duet appears there, keep the flow on whatever model you already trust. ## Claude Duet cost and speed against current options Budget against the models you already run, whose numbers are public, because Anthropic hasn't published a separate Duet price or speed; Duet uses your plan's usage limits. ![Output speed of Claude models, including Duet's model](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/174fd264-ac09-43ee-92e1-377cebe28413/claude-duet-features-cost-and-workflow-setup-in-31d960db.svg "Source: modelgrep.com") ### Claude Duet pricing and plan costs Anthropic's pricing page lists Pro at $17 a month billed annually, or $20 billed monthly, making the annual plan $36 cheaper per year, a difference worth committing upfront to secure, which means the savings are only realized if you maintain the subscription for the full year. A ten-person team can total its seats from the table before choosing, which shows the plan prices we can confirm and leaves Duet's row marked unpublished. | Plan | Monthly price | |---|---| | Claude Pro | $17 annual / $20 monthly | | Claude Max 5x | $100 | | Claude Max 20x | $200 | | ChatGPT Go | $8 | | ChatGPT Plus | $20 | | Gemini AI Plus | $4.99 | | Duet | Included in Pro and Max at no extra cost | Where workflows differ is per-token cost. [AgentsCamp](https://agentscamp.com/guides/advanced/llm-api-pricing-2026) puts Sonnet 5.5 at $2 input and $10 output per million tokens. In the same comparison, the older GPT-4o was listed at $2.50 and $10, so it saved nothing on output. ![Input and output token prices across current AI models](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/a1a8543e-e2d2-4af0-8eaf-1c2a92f2d9d0/claude-duet-features-cost-and-workflow-setup-in-6319ddef.svg "Source: AgentsCamp (2026)") [Gemini](https://gemini.google/subscriptions) 3.1 Pro is $2 and $12, which makes output-heavy jobs 20% dearer. Gemini 2.5 Pro, at $1.25 and $5, halves the output bill, which means output-heavy workloads could save substantially on API costs. ### Speed of the model behind Duet Don't assume the flagship is the slow option: [Modelgrep](https://modelgrep.com/best/fastest/anthropic) measures Opus 5 at 222 tokens per second and Opus 5.5 at 221. Older Haiku 4.5 measured 191 and Opus 4.8 169. Sonnet 5.5 runs at 111, ahead of Sonnet 5 at 94, so 5.5 delivers responses faster. A 2,000-token reply takes about 18 seconds on Sonnet 5.5 and about 9 on Opus 5.5, which decides where a timeout belongs, meaning you can't safely use the same timeout for both models. Until Duet's output speed appears, don't set timeouts around it, and remember that it draws on your plan's usage limits. Its underlying model, Opus 5.5, is published, but its speed is not. ## Next changes and weekly actions Spend Monday on one small, approved task instead of a rollout, and watch the maker's own pages for the numbers that are missing. Nothing here predicts what the maker will ship; it only tells you where changes will show up first. ### Claude Duet pages to re-check weekly Re-check four pages, because each answers a different question. The pricing page tells you whether a budget line can exist yet. The changelog tells you whether behavior changed since you last tested, which matters if a teammate built a habit around the old behavior. Whether the code you would depend on is actively maintained shows up in the GitHub repository, and the open issues are often the first honest list of rough edges. The integration listing for the automation platform tells you whether you can connect Duet without writing glue code yourself. Put a recurring slot on one person's calendar, say fifteen minutes at the Monday standup. If nobody owns the check, nobody does it. Keep the commitment small and the evidence real by working in the order below. Each step produces something your team can read at the next standup. 1. Join the Duet waitlist from a Pro or Max account, since those are the plans the maker's pages tie access to. 2. Re-check the pricing, changelog, GitHub and integration pages, and note what changed. 3. Pick one low-risk task, such as a draft nobody sends without review. 4. Run it with human approval on every output. 5. Record what happened: what it did, what you corrected, and how long review took. Bring that record to the next decision, and keep it honest and short. ## Frequently asked questions Is Claude Duet free to use? It is included in Pro and Max plans at no extra cost during the research preview. Pricing is stated only on the maker's own pages, and until one of them names a free tier, plan and budget as if usage costs money. Check the pricing page before you promise a free pilot to your team. Who made Claude Duet? Anthropic, the company behind the Claude models, is the maker implied by the product name. Confirm it on the product page itself. Anything a third-party blog says about ownership, licensing, or roadmap is secondhand until the maker repeats it. ### How do I verify integration support? No maker page confirms a native connection, so treat it as unverified. Before anyone builds a process around it, ask whether the maker documents an integration or an interface your automation tool can call. If neither exists, plan a small test first and keep the old process running alongside it. That way a missing connector costs you an afternoon, not a week of broken handoffs. ![Old bridge in use beside an unfinished new bridge](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/aa24ad19-5bbd-4208-8dd6-4c980c17f595/claude-duet-features-cost-and-workflow-setup-in-6f94cd8a.webp) Where do I find the official Claude Duet documentation? Start at the maker's own documentation site and the Claude Duet product page, and read nothing else until those have answered your question. Bookmark both and note the date you read them. When a teammate quotes a feature in standup, the first question is whether it appears on one of those two pages. ## Related reading - [Twilio MCP Server Setup for Claude in 2026](https://www.activepieces.com/blog/twilio-mcp-server-setup-for-claude-and-ai-agents-2026) - [Zendesk MCP Server Setup Guide for Claude (2026)](https://www.activepieces.com/blog/zendesk-mcp-server-setup-guide-for-claude-2026) - [Granola MCP Server Setup for Claude, ChatGPT, Cursor](https://www.activepieces.com/blog/granola-mcp-server-setup-for-claude-chatgpt-cursor) ## References - [AgentsCamp](https://agentscamp.com/guides/advanced/llm-api-pricing-2026) - [modelgrep.com](https://modelgrep.com/best/fastest/anthropic) - [Claude](https://claude.com/pricing)