Microsoft Enhances Copilot Cowork with Flexible Pay-As-You-Go Pricing

Microsoft is changing the payment structure for its AI-driven Copilot Cowork platform by implementing usage-based pricing as it broadens access to the enterprise productivity tool and pushes for more affordable AI models to cut costs.

Moreover, in an effort to cut expenses, Microsoft is considering a refined version of DeepSeek V4 or another open-source AI model as a more budget-friendly alternative to the Anthropic and OpenAI models that currently support Copilot Cowork. The company anticipates revealing its final decision in the upcoming weeks, as reported by Axios. 

Should Microsoft proceed with DeepSeek, the model would be optional for clients and entirely hosted on Azure, ensuring that customer data remains within Microsoft’s cloud and is protected by Azure’s enterprise security, compliance, and data-residency measures.

Charles Lamanna, Microsoft's EVP for Copilot, told Axios that testing revealed Copilot Cowork cannot be offered for unlimited use. “We have users who do hundreds of tasks a week, which is great—they’re way productive—but the consequence is the costs can go very high,” Lamanna said.

Under the new structure announced yesterday, the tool still necessitates a paid Microsoft 365 Copilot subscription. Now, each task executed is charged separately based on the amount of computing power it uses.

“Copilot Cowork requires the Microsoft 365 Copilot User Subscription License (USL). Users are then billed for Cowork on a usage-based basis, with charges determined by the tasks they run,” the company said in a statement. 

Light tasks use few sources and require limited reasoning to produce one output. While medium tasks use multiple sources and structured reasoning for two or more outputs, heavy tasks aggregate broadly and generate many outputs. Microsoft has identified four user personas based on these task types.

According to Microsoft, grouping users by persona and their light, medium, and heavy tasks, and applying a cost per prompt, creates a flexible cost model. This model estimates costs by multiplying the number of users in each segment by their expected prompt volume for each task type, then summing the results.

“Before usage ramps, admins can set spending limits and allocate budgets using the cost management controls described above. Tenants that had at least one user in the Frontier programme (March 30 to June 16) who used Cowork during that period get a grace period on Cowork usage and will not be billed until July 1 to support the transition,” it said. 

At general availability, Copilot Cowork uses Anthropic models, Opus 4.8 and Sonnet 4.6. In the Frontier environment, customers can access GPT 5.5, with Cowork version 1 arriving soon.

Cowork 1 will offer cost-effectiveness, high quality, and enterprise-grade performance while eliminating model bias for everyday Copilot tasks and cost-sensitive workloads.