Claude Cowork Explained: What It Is, Price and How to Get It

Claude Cowork was Anthropic's attempt to give non-developers a version of Claude Code: a desktop AI assistant that works directly inside your files and folders rather than through a chat window. It launched as a research preview in January 2026, reached general availability with enterprise controls in April, and by October 2026 it no longer exists as a separate product at all — Anthropic folded it into the main Claude app in September. If you're searching for "Claude Cowork pricing" or "Claude Cowork download," the short answer is that you now get the same capabilities simply by using Claude itself on a Pro or Max plan, starting around $17–$20 a month.

What Claude Cowork actually was

Anthropic introduced Cowork on January 12, 2026, describing it as built on "the very same foundations" as Claude Code, its command-line coding agent, but packaged for people who don't write code. Instead of typing instructions into a terminal, a Cowork user pointed Claude at a folder on their computer — a desktop folder of receipts, a shared drive of client notes, a downloads folder full of PDFs — and asked it to do something with the contents. Claude could then read, edit, create, and reorganize files inside that folder across multiple steps, without the user re-explaining the task at every turn.

The pitch was aimed squarely at non-technical knowledge workers: marketers, consultants, operations staff, and administrators who had watched developers get agentic superpowers from Claude Code and wanted something similar for spreadsheets, slide decks, and status reports. Early use cases Anthropic highlighted included turning a folder of receipt screenshots into an expense spreadsheet, tidying a chaotic downloads folder, and drafting a report from scattered meeting notes.

At launch, Cowork was a research preview restricted to Claude Max subscribers on the macOS desktop app. It was deliberately narrow: no Projects, no chat sharing, no Memory, and no syncing between devices. Those limitations didn't last long — Anthropic shipped new Cowork capabilities on a roughly biweekly cadence for the rest of the year.

How the agent worked, step by step

Cowork's model was "say what, not how." Rather than issuing a precise sequence of commands, a user described a goal — "pull the numbers from these invoices into one spreadsheet and flag anything over budget" — and Claude planned and executed the task itself, showing each step (which file it opened, which tool it used) as it worked. Anthropic built the feature to run sessions that could continue even after the user closed their laptop, and gave users the ability to step in and redirect the agent mid-task if it went down the wrong path.

Over 2026, Anthropic layered in more of the agentic plumbing that had previously been exclusive to developer tools. A plugin marketplace arrived in late February, followed immediately by recurring task scheduling and a "Customize" panel for managing skills, plugins, and connectors. In March, Anthropic added a research preview of computer use through a feature called Dispatch, letting Claude open applications and navigate a screen directly rather than being limited to file operations — and added persistent, remotely managed task threads so a Cowork job kicked off on desktop could be checked on from a phone.

The feature set that defined Cowork

By the time Cowork reached general availability, its toolkit included:

• Folder-scoped file access — reading, creating, and editing files inside a folder the user explicitly grants access to.
• Multi-step, persistent tasks — agent sessions that keep running and can be checked on or redirected later rather than requiring a single back-and-forth chat turn.
• Scheduled and recurring tasks — daily, weekly, or monthly jobs set up once and repeated automatically.
• Computer use via Dispatch — Claude operating applications and navigating a screen directly, introduced in research preview in March 2026.
• Connectors — integrations with services such as Google Drive, Slack, Microsoft 365, Amplitude, and (after April) Zoom via the Model Context Protocol.
• A plugin marketplace — packaged bundles such as Brand Voice, Legal, and Finance plugins that adjust how Claude writes and formats output for a given domain. The site has a separate explainer on how that marketplace works if you want the full mechanics.
• Custom Skills — reusable, packaged instructions for repeatable work, covered in more depth in this site's guide to Claude's Skills system.
• Sub-agents — the ability to spin up specialized helper agents for discrete parts of a larger task.
• A built-in browser — added in late August 2026, letting Claude navigate the web, fill out forms, and read pages using an isolated browser instance that never touches the user's own tabs, bookmarks, or saved passwords. Banking, email, and single sign-on sites were excluded by default unless a user explicitly enabled them.
• Memory — shared across chat and Cowork sessions from late August 2026, with editable topics and sensitive-content settings.

Claude Cowork vs. Claude Code

The comparison search queries ("claude cowork vs claude code") point at a real source of confusion, because Anthropic built both tools on the same underlying agent architecture. The difference was always about audience and interface, not capability depth:

Claude Code remains a separate, standalone product: a terminal-first coding agent that developers install via the CLI, a VS Code or JetBrains extension, or a desktop app, and that works against a codebase rather than a general-purpose folder. It wasn't absorbed into "just Claude" the way Cowork was. If you're deciding which agent fits a technical workflow, this site's comparison of Claude Code against Cursor, GitHub Copilot, and Codex covers that side of the market in more detail.

Platforms and availability through 2026

Cowork's platform support expanded in stages. It began macOS-only in January 2026. Windows support followed as part of the April 9 general-availability update, putting Cowork on both macOS and Windows through the Claude desktop app. In July 2026, Anthropic extended it beyond desktop to web and a mobile beta for Max plan subscribers, adding "remote sessions" that saved work to a user's Claude account so it could be picked up on a different device. The built-in browser feature that followed in August added Linux support, initially in beta, alongside macOS and Windows.

What Claude Cowork actually costs

Anthropic never sold Cowork as a separate paid product with its own price tag. It was gated by subscription tier, and that remains true now that it's simply part of Claude. According to Claude's current pricing page, the plans that unlock the capabilities formerly branded as Cowork are:

That last column matters if you're specifically hunting for "Claude Cowork pricing": as of Anthropic's own pricing page, the agentic capabilities that used to carry the Cowork name are described as "rolling out to Pro and Max, with more plans to follow" — meaning Team and Free users were, as of this writing, still waiting on the full rollout. There is no Cowork-specific add-on fee; you pay for a Claude plan, and the capability comes with the tier.

One practical wrinkle worth flagging honestly: agentic, multi-step work of the kind Cowork specialized in tends to consume a plan's usage allowance much faster than ordinary back-and-forth chat, because each step in a long task is itself a separate model call. Anthropic's pricing page doesn't publish an exact multiplier, so the safest approach if you're on the $20/month Pro tier and planning to run long unattended tasks is to monitor your usage rather than assume chat-level consumption.

How to get it today

Because Cowork is no longer a distinct app, there's nothing separate to search for and download. You get the same functionality by installing or opening Claude itself:

• Desktop: download the Claude app for macOS, Windows (x64 or ARM64), ChromeOS, or Linux from Anthropic's site.
• Web: use Claude directly at claude.ai in a browser, no install required.
• Mobile: the Claude app for iOS and Android, with the unified Cowork-and-chat experience rolling out there as part of the same September 2026 update.

Any of those surfaces requires a paid Pro or Max subscription to unlock the agentic, multi-step task features that used to require the separate Cowork app. Free-tier users can use standard Claude chat but don't currently get the full task-automation toolkit. Enterprise and Team admins should expect a staged rollout controlled from their organization's admin console rather than an instant switch-on.

The enterprise push that preceded the merger

Before Anthropic folded Cowork into Claude outright, it spent the spring building out an enterprise feature set clearly aimed at IT and security teams evaluating the tool for company-wide rollout. The April 9, 2026 general-availability announcement bundled six additions: role-based access controls that let Enterprise admins group users (manually or via SCIM) and assign permissions; per-team spend limits set from the admin console; usage analytics surfaced in both the admin dashboard and the Analytics API, tracking sessions, active users, and per-skill or per-connector invocation counts; expanded OpenTelemetry support emitting events for tool calls, file operations, and approval status, compatible with SIEM pipelines like Splunk and Cribl; a Zoom connector built on the Model Context Protocol delivering meeting summaries and transcripts into Cowork workflows; and the ability for admins to restrict specific actions within any MCP connector — for example, forcing read-only access organization-wide.

That enterprise groundwork is also why the September merger didn't simply flip a switch for every customer. Anthropic committed to giving Enterprise admins more than 30 days' notice before changing how the unified experience appeared in their organization, which is a meaningfully slower rollout than the "available now" treatment Pro and Max subscribers got.

Why Cowork stopped being a separate product

On September 16, 2026, Anthropic announced that "Claude Cowork and chat are now one Claude." The stated reason was friction: users had to decide upfront whether a task belonged in the chat interface or the Cowork interface, and switching between the two meant losing context partway through a piece of work. A quick question that turned into a multi-step project — draft me a report, now turn it into slides, now schedule it to re-run weekly — previously meant leaving one app for another. Anthropic's fix was to stop making users choose: any conversation can now scale from a simple question into a long-running agentic task, a document, or a presentation without switching surfaces.

The merger shipped alongside two new tools, not just a rebrand. Claude Docs lets a user co-write a document inside a conversation, edit it directly, leave comments for Claude to act on, and share it via a persistent link that works the same on mobile and desktop; it can also be exported to Google Docs or Microsoft Word. Claude Slides does the same for presentations — generate a deck from a conversation, revise it, present it directly from Claude, or download it as PowerPoint or PDF. Claude Design, previously a separate product, was folded into the same unified experience for existing users.

The rollout order mirrored the rest of Cowork's 2026 history: Pro and Max subscribers on web, desktop, and mobile got it first, over a multi-week window, with Team and Free plans explicitly described as following later.

What to watch out for

Anthropic's own documentation for agentic file and computer-use features has consistently flagged the same category of risk since Cowork's earliest research-preview days: an agent that can read, edit, and delete files — or operate a browser — on a user's behalf can act on bad instructions or manipulated content just as readily as good ones, including so-called prompt-injection attempts buried in a file or web page the agent is asked to process. The practical mitigations Anthropic has built in include scoping file access to an explicitly granted folder rather than the whole filesystem, excluding sensitive site categories like banking and email from the built-in browser by default, and showing each step an agent takes so a user can interrupt a task that's gone off course. None of that makes the risk zero, and it's worth treating any instruction to "review this file and act on what it says" with the same skepticism you'd apply to an email attachment from an unknown sender.

What's next for Cowork's capabilities inside Claude

With the Cowork brand retired, the relevant question going forward isn't "when does Cowork get a new feature" but "when does Claude." Based on Anthropic's own stated rollout plans, the near-term items to watch are the extension of the unified chat-plus-agent experience to Team and Free plans, continued expansion of the plugin marketplace and connector catalog that started under the Cowork name, and broader availability of the computer-use and built-in-browser capabilities beyond their initial beta platforms. Anthropic has also said Enterprise organizations will see the merged experience on a delayed, admin-controlled schedule rather than all at once, so larger companies evaluating this for compliance-sensitive teams should expect a longer runway than individual Pro or Max subscribers got.

Written byTheo Park

Theo Park runs the AI desk at Pandromeda. He follows model launches from the frontier labs and the open-weight community, tracks the assistants and developer tools built on them, and explains what each release changes on pricing, capability and safety. His reporting leans on primary sources: model cards, technical reports, API documentation and the companies' own announcements.

More from Theo ParkAll AI coverage