Microsoft said Tuesday that Copilot Cowork is now generally available worldwide, moving the enterprise AI agent from preview into full release with a new pricing model tied to usage. The company is positioning the product as a tool for long-running, multi-step work that can run across multiple systems and return a completed result rather than a simple draft.
The launch comes after three months in Microsoft’s Frontier preview program. Microsoft said more than half of the Fortune 500 have used the product during the preview, alongside customers including Accenture, Avanade, Advance Local, Capital Group, Koch, LTM, Ooredoo Qatar and Zurich Insurance.
Microsoft describes Copilot Cowork as an agentic system for complex tasks that may involve several tools, large amounts of context and extended runtime. In examples cited by the company, teams used it to automate spreadsheet changes and create dependency charts, compare thousands of files across product versions and analyze stalled sales pipelines to surface at-risk deals.
Copilot Cowork requires a Microsoft 365 Copilot User Subscription License. After that, customers are billed for Cowork on a usage basis, with charges determined by the tasks they run. Microsoft says pricing is measured in Copilot Credits and reflects several factors, including model use, context retrieval, tool calls and runtime.
The company said customers can choose between pay-as-you-go billing and a pre-committed volume option. PayGo is priced at $0.01 per Copilot Credit. Microsoft also plans to let users see pricing at the task level in credits, a feature it says is coming soon after GA.
Microsoft added that customers can model expected costs based on task types it observed during the preview, which it grouped into light, medium and heavy work. The company also published a spreadsheet to help estimate costs, and said the estimates assume Anthropic Opus 4.8.
At general availability, Copilot Cowork runs on Anthropic models, including Opus 4.8 and Sonnet 4.6. Microsoft said customers in Frontier can also use GPT-5.5, and a new model called Cowork 1 is expected in the coming weeks. The company said Cowork 1 will be tuned for lower-cost enterprise use.
Because usage-based billing can make spending harder to predict, Microsoft is adding a set of cost controls. Admins can keep Cowork turned off by default, decide who gets access and set spending limits at the tenant, group and user levels. The company also said customers can configure alerts when spending reaches specific thresholds and allow users to request additional credits from within Cowork.
Microsoft said customers will also get usage reporting broken down by tenant, group and user, with task-level pricing visibility coming soon. The company framed the new controls as part of a broader effort to help organizations monitor return on investment as agentic AI becomes more compute-intensive.
There are also enterprise security and compliance features built in. Microsoft said prompts, responses and generated outputs remain within Microsoft 365 controls, with support for audit logs, data security posture management, eDiscovery, insider risk management, data lifecycle management and communication compliance. Data loss prevention support is listed as coming soon.
The Microsoft 365 Copilot app now includes a toggle that opens the full Cowork experience, making it easier to move from chat to action. Microsoft also said nine partner plugins are available now, with eight more on the way, along with support for the Dynamics 365 suite and Fabric. Browser use through Edge is available in Frontier, and the company said it follows existing enterprise policies.
Microsoft said billing for Copilot Cowork starts immediately, but some customers will get a grace period. Tenants that had at least one Frontier user who used Cowork between March 30 and June 16 will not be billed until July 1, 2026.
The general availability release underscores Microsoft’s push to make Copilot a broader platform for workplace automation, not just a chat assistant. With Cowork, the company is betting that enterprise customers will pay for AI that can carry out more of the work on their behalf, provided they can keep a close enough handle on cost and governance.