Understand Cowork usage and cost

Cowork does real work for you—reading files, drafting documents, searching the web, generating images, and carrying out multi-step tasks across your apps. That work runs on computing resources, and like any cloud service, it has a cost behind it.

You don't need to do math before you type. You do need to ask deliberately—the same way you'd think for a second before kicking off a huge print job.

Cowork usage is pay-as-you-go

Cowork uses a pay-as-you-go model. Instead of a flat, all-you-can-use license, the work Cowork does for you counts toward your organization's consumption, measured in Copilot Credits.

Credits add up while Cowork is actively working:

• Generating responses and drafting content
• Calling tools and skills to complete steps
• Creating images
• Running browser tasks on the web

Quick questions sit at the low end. Multi-step projects—researching several sources, building a document, drafting follow-up messages—sit at the high end, because Cowork is doing far more.

Your admin sets your limit

Your administrator sets the credit limit for your organization. If you reach it, you request more, your admin reviews the request, and you pick up where you left off once it's approved.

That's the practical reason to pay attention. Nothing breaks, and nothing lands on you personally—you'd just pause until your admin approves more.

Check a task's usage

Type /cost in a conversation and Cowork shows how many credits the current task has used so far—for example, "610.6 credits used for this task so far." Checking it now and then builds your intuition for which requests are light and which are heavy, so you can right-size how you ask.

The model you choose changes usage

Cowork can run on more than one AI model, and you pick one from the model selector. Models differ in how deeply they reason and how much work they do per turn, which affects usage along with the speed and depth of the response.

For most work, leave the selector on Auto. Auto matches the model to the task you describe, so a quick request uses a lighter model and a complex one gets more capability—without you having to choose. Reaching for the most powerful model on every small task uses more than the work calls for.

To learn how the models differ and when to change the selection, see Choose a model for Copilot Cowork.

Work with usage in mind

Working efficiently isn't about asking Cowork for less. It's about getting to a good result without extra laps:

• Be clear and specific. A well-scoped request reaches a good result with less back-and-forth.
• Hand off the whole goal. Cowork is built for multi-step work, so describe the outcome you want rather than feeding it one small step at a time.
• Stay on Auto unless you have a reason to switch. It balances capability and usage for you.
• Check /cost occasionally. It's the easiest way to learn what your typical work uses.