Microsoft made Copilot Cowork generally available on 16 June 2026, and it runs on Anthropic's Claude models. So the choice between Copilot Cowork and Claude Cowork on 3P is not a choice of model. It is a choice of where the agent runs, what it can reach, and how you pay for it.
Both products take a described outcome, plan the steps, work across tools and return a finished result rather than a draft. For a buyer in a security review, the questions that separate them are architectural: whose account the inference lands in, whose logs record it, and whether the bill scales with seats or with work.
Copilot Cowork runs inside Microsoft 365, grounded in Microsoft Graph, and costs a 30 dollar per user Copilot seat plus metered Copilot Credits. Claude Cowork on 3P runs the same Claude models through Amazon Bedrock, Google Cloud Vertex AI or Microsoft Foundry in your own cloud account, reaches any system through MCP connectors, and bills as cloud consumption with no seat licence. For a mixed tool stack or a regulated data boundary, Claude Cowork on 3P is the stronger fit. Moterra deploys it in one to two weeks.
Is the model actually different?
Not at the top end. Microsoft launched Copilot Cowork on Anthropic's Opus and Sonnet models, with other model families alongside and Anthropic models enabled per tenant by the admin. Claude Cowork on 3P calls the same Claude models from your cloud provider's catalogue.
When the reasoning is the same, the decision moves to the layer around it. That layer is where the two products are built in opposite directions.
Core structural differences
Where does inference run?
Copilot Cowork is cloud-only and runs inside Microsoft 365. Its controls are Microsoft's controls: permissions, sensitivity labels and Purview auditing, applied within Microsoft's service boundary. That is a strong position if Microsoft 365 is already your system of record for governance.
Claude Cowork on 3P moves the model call into your perimeter. A managed configuration, pushed by your MDM, points every laptop at an endpoint in your own cloud account. Four boundaries move with it:
For a team that has to show a regulator exactly where prompts and files are processed, that difference is usually the one that decides the review.
What can the agent reach?
Copilot Cowork is anchored in Microsoft Graph. It sees email threads, Teams conversations, calendar history and SharePoint files, and Work IQ understands how they relate. Reach beyond Microsoft 365 comes through partner plugins such as monday.com, Miro and Harvey.
Claude Cowork on 3P is anchored in the Model Context Protocol, an open standard. Any system with an MCP server can be connected, and each connector is approved one by one through an allowlist in the managed configuration. That includes Microsoft 365: connectors to SharePoint, Teams and Outlook inherit the permissions already in place, so a person sees exactly what they could already see.
The practical result is that Claude Cowork on 3P can reach the Microsoft 365 data Copilot is built around, and the Salesforce, Slack, Jira, HubSpot and Notion data it is not.
How is each priced?
Copilot Cowork stacks two costs. A Microsoft 365 Copilot licence is required, at 30 dollars per user per month, and Cowork usage is then billed through Copilot Credits at 0.01 dollars per credit on pay-as-you-go, or from prepaid packs. Task cost depends on the models, data and actions involved, and Microsoft provides spending policies and per-user caps to manage it.
For 100 users, that is 3,000 dollars a month in seats before the first credit is spent.
Claude Cowork on 3P has one variable cost for the model: consumption, billed by your cloud provider, on the account you already pay. There is no seat floor, so a user who runs one task a week costs one task a week. On AWS, consumption can be procured through AWS Marketplace and counted against existing commitments.
When does Copilot Cowork fit?
Copilot Cowork is the simpler choice when nearly all work happens inside Microsoft 365, governance already runs through Purview, Copilot seats are already budgeted, and long tasks need to run unattended in Microsoft's cloud.
Claude Cowork on 3P is the stronger choice when any of these hold:
How does Moterra deploy Claude Cowork on 3P?
Private Claude Cowork by Moterra is Claude Cowork on 3P, deployed and run inside your own AWS or Azure account. Same Anthropic product, same models, same interface. What changes is where it runs.
Moterra is a Select Services Partner in the Claude Partner Network and an AWS Partner with GenAI Competency, ISO 27001 and ISO 42001 certified. The mechanics of the route are covered in our guide to Claude Desktop on 3P, and the deployment around it in Claude Cowork for business.
The quote configurator sets Year 1 cost side by side with Microsoft 365 Copilot, so the comparison above can go to finance as a number.
See Claude Cowork on 3P against your Copilot shortlist.
Tell us where you are and a Moterra specialist will confirm a time within one working day.
Goes straight to your sales team · no marketing list · no follow-up sequence
Frequently asked questions
Do Claude Cowork on 3P and Copilot Cowork use the same model?
Largely, yes. Copilot Cowork launched on Anthropic's Claude models, Opus and Sonnet, alongside other model families. Claude Cowork on 3P calls the same Claude models through Amazon Bedrock, Google Cloud Vertex AI or Microsoft Foundry. The decision is about deployment, data reach and cost, not model quality.
Can Claude Cowork on 3P work with Microsoft 365 data?
Yes. MCP connectors to SharePoint, Teams and Outlook read and write while inheriting the permissions already in place, so a person sees exactly what they could already see. Moterra deploys those connectors alongside connectors to Salesforce, Slack, Jira and Google Workspace.
How is Copilot Cowork priced?
Copilot Cowork requires a Microsoft 365 Copilot licence, listed at 30 dollars per user per month, and bills usage on top through Copilot Credits at 0.01 dollars per credit on pay-as-you-go, or through prepaid credit packs. The cost of a task depends on the models, data and actions it uses.
How is Claude Cowork on 3P priced?
Model usage is billed as consumption by your cloud provider, so it lands on your existing AWS, Google Cloud or Azure bill and can count against committed spend. There is no per-seat licence for the model. With Moterra, the managed service is priced separately in the open quote configurator.
Where does inference run in each product?
Copilot Cowork runs in the Microsoft 365 cloud, inside Microsoft's service boundary. Claude Cowork on 3P sends model calls to an endpoint in your own cloud account, in the region you choose, logged in CloudTrail or Azure Monitor and encrypted with your own keys.
When does Copilot Cowork make more sense?
When nearly all work lives in Microsoft 365, governance is already run through Purview, and the organisation has budgeted for Copilot seats. Teams with a mixed stack, a data residency requirement tied to their own account, or no wish to buy seats on top of consumption usually fit Claude Cowork on 3P better.
How long does a Moterra deployment of Claude Cowork on 3P take?
Typically one to two weeks to the first users, starting with one department and one workflow, deployed into your AWS or Azure account and managed through your existing MDM.
