Claude Cowork vs Copilot Cowork (and ChatGPT Work): which fits your work?

TL;DR verdict box

Choose Claude Cowork for a defined folder of work. Choose Copilot Cowork when the job depends on your Microsoft 365 tenant context. Choose ChatGPT Work when you want OpenAI’s cowork-style surface across desktop, web and mobile, with plugins and scheduled tasks on the plans that include them. Start with where the source material lives, then compare the cost and controls needed to delegate the task.

One naming change matters: Anthropic's pages now say Claude Cowork is becoming “Claude”, rolling out to Pro and Max first. I keep the Cowork name here so the comparison stays clear. See the Claude product page. OpenAI’s cowork-style surface is labelled ChatGPT Work on the tracker; ChatGPT agent is no longer the product to look for (see the FAQ below).

Side-by-side table

The tracker provides the starting comparison below. Official pages qualify its pricing and capabilities; Craft ratings are editorial judgements.

Decision	Claude Cowork	Copilot Cowork	ChatGPT Work
Starting workspace	Folders and tools you choose	Microsoft 365 work context	Desktop, web and mobile ChatGPT surfaces (Free/Go limited to desktop)
File relationship	Works directly on selected files	Works across tenant content and attached files	File uploads; code edits on macOS on Go+
Entry price, USD	Pro: $20 monthly; displayed $17/month annually, $200 upfront	Microsoft 365 Copilot: $30/user/month, paid yearly, plus a qualifying Microsoft 365 plan	Dollar prices not shown on the pricing page; limited on Free/Go
Additional cost structure	Subscription with usage limits	Cowork usage metered in Copilot Credits; PayGo $0.01/credit	Plan tier (Plus/Pro) and Business/Enterprise for admin controls
Connections	Connectors and plugins	Work IQ, connectors and plugins	Plugins on all plans; apps connecting to internal tools + Developer mode (beta) on Plus/Pro
Governance emphasis	Folder scope; Enterprise role-based access; Team/Enterprise telemetry	Tenant permissions, Purview and audit; spending policies	Business/Enterprise only
Scheduling	Supported	Supported	Scheduled tasks on Go+
Craft fit	Augment / Adapt	Augment	Augment

Prices come from Claude pricing, Microsoft enterprise pricing, Microsoft's Cowork launch announcement and ChatGPT pricing. These are US-page figures where shown; ChatGPT’s pricing page does not list dollar amounts for Work on the vendor pages we checked.

Where it works

Claude's useful starting point is a selected folder: exports, notes, reference documents and a deliverable to produce. Anthropic describes direct work in folders and tools, with web and mobile rollout also listed. Local-file work remains a useful distinction, but “desktop only” would now be misleading.

Copilot's starting point is the organisation's working context. Microsoft describes Work IQ grounding across emails, meetings, files and business data. Its product page positions Cowork as work spanning apps and systems.

ChatGPT Work’s starting point is the ChatGPT surface itself: desktop, web and mobile on Plus and Pro, with Free and Go limited to desktop. Official pricing describes file uploads, plugins on all plans, apps that connect to internal tools, and Developer mode (beta) on Plus and Pro. Code edits on macOS appear on Go and above. Use that when the work lives in ChatGPT’s file and plugin model rather than a Claude folder or an M365 tenant.

Imagine a monthly operating review. If you already have the approved inputs collected, trial a bounded file task in Claude. If the hard part is finding the latest decisions scattered through company conversations, trial a tenant-context task in Copilot. If the team already works in ChatGPT with plugins and scheduled tasks, trial Work on a representative package of uploads. In every case, ask for source references and unresolved gaps alongside the finished report.

Pricing

Claude Pro has the simpler entry cost: a flat subscription, subject to usage limits. Anthropic lists $20 when billed monthly, or a displayed $17 per month with an annual subscription costing $200 upfront. Do not multiply the rounded monthly display to infer the annual invoice. Cowork is available on all paid plans, according to Anthropic's enterprise announcement.

Microsoft lists Microsoft 365 Copilot at $30 per user per month, paid yearly, with a separate qualifying Microsoft 365 licence required. Its Cowork launch announcement says that licence is required and Cowork consumption is additional, denominated in Copilot Credits. PayGo costs $0.01 per credit. A credit is a billing unit, not a promise of one completed task.

ChatGPT Work is listed on ChatGPT pricing without dollar amounts on the vendor pages we checked: limited on Free and Go (desktop only), and available on Plus and Pro across desktop, web and mobile. Admin controls are Business and Enterprise only.

For an existing Copilot customer, compare the extra Cowork consumption with the cost of adding Claude or ChatGPT. For a new customer, include the qualifying Microsoft 365 subscription and Copilot licence, or the ChatGPT plan that actually unlocks Work on the surfaces you need. Comparing only Claude Pro with a credit rate or an unpriced ChatGPT row hides most of the purchasing decision.

Run a small pilot using representative work. Record consumption, completion quality and review time. Repeat the task with realistic inputs before budgeting a wider rollout. A simple task can be cheap yet unhelpful; a costly task can still pay for itself if it saves substantial preparation and checking.

Governance & approvals

Microsoft is the stronger starting point when the organisation already governs work through Microsoft 365. Its governance guide documents Purview, audit visibility and execution under user permissions.

Access has a specific trap: a spending policy selecting Cowork grants access to users in its scope. A tiny budget does not block access. Excluding users from those policies is the relevant control. Separately, disabling the Anthropic model family does not disable Cowork itself.

Claude offers a different boundary. You select accessible folders and tools; deletions require approval. Its enterprise announcement specifies role-based access for Enterprise and OpenTelemetry for Team and Enterprise. Telemetry covers file activity, tool calls and approval events. Do not assume a personal Pro subscription includes those organisational controls.

ChatGPT Work puts admin controls on Business and Enterprise only. Individual Plus and Pro users still need a human check before output goes to anyone else; organisational policy lives on the business tiers.

My judgement: neither product earns a blanket “safer” label. Match the controls to the action. Reading approved material, editing a shared document and sending an external message need different delegation rules. Define the permitted sources, destination, approval point and recovery process before handing over the task.

Connections

Claude uses connectors and plugins. Anthropic describes plugins as bundles of skills, connectors and sub-agents. Think of a connector as access and a skill as a procedure: access to a system does not explain your organisation's preferred way of doing the work.

Copilot combines Work IQ context with extensibility. Microsoft's FAQ confirms plugins can add skills and connectors. This is not a clean division between “Claude has plugins” and “Microsoft has company data”. Both connection design and permissions matter.

ChatGPT Work lists plugins on all plans, plus apps that connect to internal tools and Developer mode (beta) on Plus and Pro. Treat those as OpenAI’s connection surface, then confirm which apps and modes your account actually has.

Treat the Graph distinction as an architectural question to ask during evaluation: which organisational sources reach the task, and under whose identity? Confirm the exact Work IQ/Microsoft Graph relationship with your Microsoft admin. For Claude and ChatGPT, confirm which connectors or apps are enabled for the signed-in account.

Use our MCP explainer for the connection vocabulary. Then evaluate an actual workflow: can the agent retrieve the approved source, preserve its meaning and produce something the next person can use? A long integration list cannot answer that.

Craft mode fit

The tracker rates Claude Cowork for Augment and Adapt, Copilot Cowork for Assist and Augment, and ChatGPT Work for Augment. These are editorial starting points, not vendor certifications or limits on capability.

In Assist, AI helps me think; I decide. Use this mode to frame a question, challenge an interpretation or decide what evidence is missing.

In Augment, AI carries several steps; I review. A source-backed briefing or assembled report is a useful trial. Specify what a successful result must contain and reserve time to check it.

In Adapt, AI works on files within rules I set; I stay on call. See Claude Chat vs Cowork vs Claude Code for the broader distinction.

Apply the seven-layer map: model, harness, connections, skills, product, agency, my role. A shared model does not make three products interchangeable. The harness coordinates work; connections expose context; skills guide procedure. Product controls shape agency, and my role determines what I must still inspect and decide.

Who should pick which

Independent consultants and analysts: start with Claude when your assignment arrives as an approved set of files. Define the output, excluded material and checks for completion. Keep the first task small enough that you can inspect every substantive claim.

Managers and Microsoft 365 teams: start with Copilot when gathering internal context is the main burden. Test whether the result finds the right decisions and distinguishes current information from superseded discussion.

Teams already on ChatGPT: start with Work when uploads, plugins and scheduled tasks on Go+ match how you already collaborate. Confirm Business or Enterprise if you need admin controls.

Operations and finance leaders: choose by evidence flow. Ask for reconciliations, exceptions and missing inputs. Treat a confident narrative without a traceable calculation as unfinished work, whichever product produced it.

IT and governance teams: evaluate the workflow alongside identity, permissions and reporting. Start with a bounded group and an explicit review owner. Expand after you understand consumption and failure patterns.

Using more than one can make sense: one task gathers approved tenant context; another transforms an authorised file package; a third runs inside ChatGPT with plugins. Decide whether each hand-off is permitted and worth the extra effort. See the Claude Cowork profile, Copilot Cowork profile and ChatGPT Work profile for the tracker entries.

FAQ

What's the difference between Claude Cowork, Copilot Cowork and ChatGPT Work?

The main difference is where you start: selected files and folders for Claude, Microsoft 365 tenant context for Copilot, or ChatGPT’s desktop/web/mobile Work surface (with Free/Go limited to desktop) for OpenAI. All three extend through connected tools. Choose by the sources needed for your task, then test whether the output is accurate, traceable and straightforward to review.

Where did ChatGPT agent go?

ChatGPT agent is no longer available. OpenAI’s current cowork-style surface for this comparison is ChatGPT Work. See ChatGPT agent and ChatGPT Work and Codex for OpenAI’s help articles.

Which is cheaper?

Claude Pro has the lower listed entry subscription among the priced rows here: $20 monthly or $200 annually. Microsoft 365 Copilot lists $30 per user monthly, paid yearly, plus a qualifying plan and Cowork consumption. ChatGPT Work’s dollar prices are not shown on the pricing page we use in the tracker. Existing licences, task volume and review effort determine the practical comparison; the headline fee alone cannot settle it.

Does Copilot Cowork use Claude?

Yes, Copilot Cowork supports Anthropic models alongside other providers. Microsoft's current model guide lists Opus 5, Claude Sonnet 5 and Claude Fable 5.1. Admins can disable the Anthropic family.

Which is safer or better governed?

Copilot is the natural starting point for organisations already using Microsoft 365 governance, with tenant permissions, Purview and audit controls. Claude offers folder scoping and organisational controls on relevant plans. ChatGPT Work’s admin controls are Business and Enterprise only. Safety depends on your configuration, connected systems, approval rules and reviewer, so test the intended workflow rather than ranking brands.

Can I use more than one?

Yes, when each has a clear job and your organisation permits the data flow. You might prepare a tenant-based briefing in Copilot, use Claude for an approved file transformation, or run a plugin-backed package in ChatGPT Work. Keep the hand-off explicit, avoid unnecessary copies and assign a person to check the final result against its sources.

Which Craft mode does each suit?

Our tracker places Claude Cowork in Augment and Adapt, Copilot Cowork in Assist and Augment, and ChatGPT Work in Augment. Treat those as editorial guidance. Start with AI helping you think, move to reviewing several completed steps, and delegate file work only when you can define rules and stay available for exceptions.

As-of + sources

Checked as of 17 Oct 2026. Microsoft confirms general availability on 16 June 2026 in its launch announcement. Product names, models and commercial terms can change; follow What changed for updates.

Not sure how much to hand over? Craft teaches Assist → Augment → Adapt on your own work. For ongoing coverage, subscribe to The Helix Loop to get Agent Watch every week, or follow What changed.

Before you rely on this

• Microsoft's model and governance guides disagree about whether Anthropic enablement can be scoped to groups. This article makes no claim about that scope; confirm it in your tenant.
• ChatGPT Work dollar prices are not shown on the pricing page we checked.
• Craft mode ratings are editorial judgement, not vendor claims.