Copilot Cowork Pricing Explained: Credits, Costs, and How to Budget Before You Enable It

The most frequently asked question among IT managers at TrustedTech at the moment isn't "What can Cowork do?" It’s this: "How can I budget for it without facing a surprise at the end of the month?" This is a valid concern. Copilot Cowork became generally available on June 16, 2026, featuring a consumption-based pricing model which is added on top of the current per-seat license, and this combination represents genuinely new ground for most Microsoft customers.

As a Microsoft CSP Direct Bill and Solutions Partner, one of the things TrustedTech does is to assist customers in estimating their licensing costs prior to receiving the bill. This article deals with that situation in the case of Cowork: it explains how the billing model works, what specific tasks have real-dollar costs, how to forecast spending according to user type, and the admin controls which stop a credit bill from exceeding the value it is supposed to represent.

All the figures mentioned herein are taken from Microsoft’s official Copilot Credits Guide (June 2026) and the GA announcement, and in cases where a figure is a planned estimate rather than a guaranteed rate this is clearly stated.

The Short Version

Copilot Cowork pricing has two parts: the M365 Copilot USL at $30/user/month, which does not change when you enable Cowork, and consumption billing in Copilot Credits at $0.01 each, which scales with the complexity and frequency of the tasks your users run. Light tasks cost under $2. Medium tasks cost $4–$6. Heavy tasks cost $15 and above. The four drivers of each task’s credit cost are the model used, the context retrieved, the tool calls made, and the runtime.

Budget by sizing your user population into Microsoft’s four personas, estimating prompt volume and task tier for each group, and running those numbers through Microsoft’s cost estimator spreadsheet. Start on pay-as-you-go, set spending controls before you enable access, and treat the first 30 days as calibration data. Move to prepaid only after you have real consumption figures to size the commitment.

If your organization is in the Frontier program and has not yet configured billing, the grace period ends July 1, 2026. If you want help modeling Cowork costs against your actual user mix before you commit any credits, TrustedTech’s team does exactly that kind of pre-deployment licensing review.

The Two-Layer Cost Model

Misunderstanding this structure is the reason why all subsequent budget discussions come to a halt. Cowork does not charge a single price; rather, it has two distinct cost layers which must be understood and budgeted on their own.

Layer 1: The Microsoft 365 Copilot license

Each user who wishes to use Cowork must have an active Microsoft 365 Copilot User Subscription License (USL). This is the standard M365 Copilot subscription at a rate of $30 per user per month and includes Copilot Chat, Copilot in Word, Excel, Outlook, Teams, and PowerPoint as well as the Work IQ grounding engine. The cost remains the same even if Cowork is enabled. If your organization is already paying for it, then that cost is already included in your budget. If you have not yet deployed Microsoft 365 Copilot, then that should be the conversation you have first. Our analysis of whether the M365 Copilot seat is worth its cost deals with that decision on its own.

Layer 2: Copilot Credits

This section is new; each task that Cowork carries out is charged for separately using Copilot Credits, a usage-based currency that is pooled at the tenant level. A single credit costs $0.01 when using the pay-as-you-go option. The M365 Copilot licence does not include any bundled credit allowances; according to Microsoft's Credits Guide, each credit that Cowork uses represents an additional cost on top of the $30 seat fee.

The credits are also split among the tenants; the same pool is used for Copilot Cowork, Copilot Studio agents, Dynamics 365 agents, Power Platform workloads, and usage of the Work IQ API. When your organization makes use of several of these services, the budget must take into account the total AI consumption, not just that of Copilot Cowork.

What Actually Drives the Credit Cost of Each Task

The section that most pricing guides omit is the reason why two organizations carrying out the "same" task can end up with very different bills. As stated in Microsoft's Credits Guide, every Cowork task consumes credits from four different inputs, and it is by understanding these four factors that the difference between a budget that stays within limits and one that unexpectedly exceeds them is made.

Models

The AI model that Cowork uses for a given task. At launch, Cowork operates using Anthropic's Opus 4.8 and Sonnet 4.6 by default, while Frontier customers have access to GPT-5.5. Carrying out more advanced reasoning on a frontier model uses up more credits than carrying out the same kind of task on a lighter model. Microsoft is set to release its own Cowork 1 model, a version that has been fine-tuned for everyday tasks and which is considerably less expensive, in the weeks after the launch. Organizations that are concerned about costs and who do not want to deploy it widely can wait until Cowork 1 is available; the credit calculations become much better for task types that are common in everyday use.

Context

The amount of organizational data that Cowork retrieves in order to carry out a task is determined as follows: it draws on emails, files, the calendar, Teams messages, and previous interactions via Work IQ. A task that has to review 90 days of email history and cross-reference three SharePoint libraries costs more in terms of context retrieval than one based on a single document. This is the reason why the quality of your M365 data environment has a direct impact on your credit bill; if you have a large SharePoint environment with poorly managed content, then Cowork has to retrieve more noise in order to locate the signal, which in turn uses up more credits and results in a lower-quality output.

Tools

The number of tool calls that Cowork makes when using various apps, such as sending an email, creating a document, scheduling a meeting, posting on Teams, and querying a plugin, adds up to a total. A task that reads files and generates a summary involves fewer tool calls than one that reads files, creates a Word document, sends it to three people, schedules a follow-up, and posts a Teams message.

Runtime

The total duration for which the orchestration runs. A task that takes three minutes costs less in terms of runtime than one that takes forty-five minutes. Since Cowork operates in the cloud and can run even when the user is not nearby their device, tasks are capable of running for anywhere from a few minutes to several hours. The greater the runtime, the higher the cloud orchestration cost, which is reflected in the total number of credits.

What Tasks Actually Cost: Light, Medium, and Heavy

The illustrative planning ranges for credits offered by Microsoft's Credits Guide are based on Anthropic Opus 4.8; however, these are not guaranteed billing rates since actual costs will differ according to workflow and usage patterns. They do represent the closest available to an official benchmark and correspond directly with Microsoft's own cost estimator spreadsheet.

Source: Microsoft Copilot Credits Guide, June 2026. The figures given are illustrative planning estimates prepared using Anthropic Opus 4.8. Actual costs will differ according to the complexity of the task, the model chosen, and the amount of data retrieved.

After carrying out the calculation, one figure stands out: a knowledge worker who performs three medium-level account preparation tasks each week uses approximately $50 to $75 per month in Copilot Credits, in addition to the $30 USL. This makes the total AI cost per user for that position amount to $80 to $105 per month. As to whether or not this figure can be justified entirely depends on how much time is currently being spent on those three tasks and on what that time is worth. The details of this kind of calculation are explored in our post on Copilot Cowork ROI by user persona.

How to Build a Monthly Budget Estimate

The three-step method for estimating Cowork spend described in Microsoft’s Credits Guide is one that TrustedTech employs when assisting customers in determining their credit budget prior to granting access.

Step 1: Size your Cowork user population by persona

It isn't necessary for every licensee of M365 Copilot to make use of Cowork on day one. The first thing to consider is which roles will actually carry out Cowork tasks frequently enough to warrant the associated consumption cost. According to data from Microsoft's Frontier programme, four different user profiles exhibited distinct usage patterns.

In most organizations, corporate knowledge workers those who use a lot of email, have to coordinate regularly and produce a lot of documents; they are by far the biggest group and generally carry out light to medium tasks. Examples include drafting weekly status updates, preparing meeting briefs, and synthesizing their inboxes.

Management and the senior leaders carry out fewer tasks, but these are of a medium or heavy nature, such as pipeline reviews, executive briefings, and cross-source analysis. The hourly rate that they achieve is the highest in the building, which means that the return on investment case is generally strong even when the rate is $10 to $15 per task.

The same type of task that customer-facing staff members carry out, such as preparing for customer meetings, synthesizing follow-up discussions, and providing account briefings, is performed repeatedly across various accounts; because of its predictable frequency, this group is the easiest to forecast.

Technical workers carry out the tasks only rarely, but at a higher level: comparisons involving large sets of files, mapping of dependencies, and the synthesis of research across hundreds of documents. It is easy to justify performing heavy tasks on a monthly or quarterly basis; daily heavy tasks, however, require careful modeling before they can be enabled on a wide scale.

Step 2: Estimate prompt volume per persona

For each persona group it is necessary to estimate the number of Cowork tasks each user carries out per week and the tier they are mostly in; a customer-facing representative who is carrying out account prep for four meetings each week is carrying out four medium tasks, a knowledge worker who automates their Monday status email and Friday prep brief is carrying out two light tasks, and a senior leader who carries out one pipeline review and two executive briefings each week is carrying out one heavy and two medium tasks.

Microsoft provides a free cost estimator spreadsheet that takes user counts and prompt mix as inputs and outputs a monthly credit estimate. It is the right tool for formalizing this into a number you can take to finance. The spreadsheet uses Opus 4.8 as its model assumption; once Cowork 1 releases, re-run the estimate with the lower model cost.

Step 3: Apply per-prompt cost and sum

To get your monthly PayGo estimate for that group, multiply the number of users in each persona group by their expected weekly prompt volume, convert that figure to a monthly amount, apply the credit range appropriate for their typical task tier, and then multiply by $0.01. To obtain the total tenant estimate, add up the amounts from all the groups.

Example for a 200-person organization with a focused initial rollout:

The following is an example based on mid-range credit estimates at PayGo prices. The actual costs will differ. These 90 users also have the $30 monthly M365 Copilot USL, which adds $2,700 monthly to the total AI cost for this group.

The entire AI cost for the 90-user Cowork pilot amounts to about $7,120 a month, comprising $2,700 in USL fees and $4,420 in credits. It is up to whether the 90 users are achieving measurable time savings in relation to workflows that cost more in labor than the amount spent on credits to decide if that figure is justifiable. This question should be addressed through a 30-day pilot before deciding to proceed with a full rollout.

PayGo vs. Prepaid: Which Option Makes Sense

Microsoft offers two ways to buy Copilot Credits, and choosing the wrong one early costs real money.

With the pay-as-you-go (PayGo) option, you are charged $0.01 for each credit, and the bill is sent at the end of the month. There is no need to make an advance commitment and usage will not be interrupted even if it exceeds the estimated amount. This is the appropriate starting point for any organization that has not yet determined its actual consumption patterns, since the ability to increase or decrease your usage without any penalty is more valuable than the discount at this stage.

The Copilot Credit Pre-Purchase Plan (P3) is a one-year prepaid credit pool which offers volume discounts. According to the Microsoft Credits Guide, the discount rates range from 5% when 300,000 credits are purchased (£2,850 rather than £3,000 under PayGo) up to 20% for 300 million credits. Although the discount is real, so is the risk since any P3 credits not used will expire at the end of the annual period; the Credits Guide makes this clear. If you make a commitment under the P3 plan for 300,000 credits and your organisation ends up using only 200,000 of them, the effective cost works out at $14.25 per 1,000 credits instead of $10, making it more expensive than PayGo.

The reasonable approach is to first run PayGo for 60 to 90 days in order to obtain accurate consumption data, and then to assess whether a commitment under any tier of P3 would result in net savings based on your actual usage and taking into account the expiry risk. TrustedTech can assist with carrying out that analysis as part of a wider Copilot licensing review. Credits acquired via either method are MACC-eligible and are available through the CSP program, so companies that already have an Azure subscription can arrange for them using their current partner arrangement.

The Frontier-to-GA Transition: What Changed and When Billing Starts

A question which TrustedTech is receiving a lot from customers at the moment is that of organisations which had been using Cowork during the Frontier preview and who are now surprised to find that it costs money. It is reasonable for them to feel that way, since Microsoft did provide a transition period, although many do not realise how short the timeline actually is.

From March 30th to June 16th, 2026, Frontier program participants received cowork at no extra cost. When the GA session took place on June 16th, usage-based billing was put into effect. Tenants who had at least one user who carried out Cowork tasks while the Frontier program was running were given a grace period, during which no billing would take place until January 1, 2026. After January 1, all Cowork usage would be measured and billed. Tenants who had Frontier users but had not set up billing by January 1st would lose access to new Cowork tasks, although their existing task history would still be accessible.

The practical consequence is that if your organization is in the Frontier region and has not already arranged billing in the M365 Admin Center, the opportunity to do so without experiencing a service interruption is shrinking. On the positive side, this same grace period is also the best time to put in place governance controls, set spending limits, and configure alerts before any credit spending starts to accumulate.

The partner action set out in Microsoft’s Partner GTM materials is clear: you must inform Frontier customers of the change, set up usage-based billing before June 30, and establish cost controls before access is opened. If you are a TrustedTech customer taking part in the Frontier program and have not already contacted us about this, then you should get in touch with our team to have the billing setup completed by the deadline.

Admin Controls: How to Govern Spend Before It Starts

The most frequent pattern that TrustedTech observes in cases where customers receive unexpected bills involves the admin enabling access and setting up governance only after usage had already taken place. The proper sequence is the opposite. Cowork is disabled by default in each tenant in order to allow for this kind of work. Below is the checklist.

Enable access deliberately, not broadly

Cowork access is set up in the Microsoft 365 Admin Center (MAC) by going to Copilot > Cost Management. Administrators have to actively turn the feature on and connect an Azure subscription for billing purposes before any user can carry out a task. You can use the discoverability switch to determine whether users are able to see or request Cowork access, regardless of whether billing has been set up. Begin by enabling the feature for only a specific pilot group rather than for the entire M365 Copilot user base.

Set spending limits before the first task runs

The MAC offers spending limits at three levels: for the entire tenant, for specific user groups (such as a department, a project team, or a single-user group when individual caps are required), and for individual users under a group policy. A reasonable initial limit for a pilot group is between $75 and $100 per user per month, as this amount gives users enough credit to make effective use of Cowork without permitting any one user to build up an uncontrolled balance. Microsoft’s Cost Management L200 documentation deals with the three most typical administrative situations when agents unexpectedly use premium services, when individual users spend without restriction, and when departments go over their allocated amount by providing a particular policy setting for each.

Configure threshold alerts

Together with the hard limits, set up usage alerts which will inform the designated recipients when expenditure exceeds a certain percentage of the established limit. The alert is triggered before the cap is reached, not after. Place the alerts at 50% and 80% of the monthly cap in order that there will be time to make adjustments before the limit is reached rather than finding out about it only after.

Review usage weekly for the first 30 days

MAC usage reporting breaks down expenditure by user, group, feature, and time period. In the first month of a Cowork pilot, the data collected is used for calibration – specifically, it shows which task types users are actually carrying out, how much those tasks cost in real credits, and where measurable time savings are appearing in the workflows? It is this data that enables a defensible decision in the second month regarding whether to expand, hold, or modify the scope, making it acceptable to finance and leadership. It would be necessary to deploy the system widely only after having gathered that data, otherwise the budget discussion becomes considerably more difficult.

One Factor That Changes Every Estimate

All budget models for Cowork are based on the assumption that the tasks carried out by Cowork do in fact generate reliable outputs that are worth acting upon; however, this assumption is only valid if the M365 data environment which underlies the Work IQ grounding is in a reasonable condition.

Cowork gets the context it needs from emails, files, the calendar, and Teams. In a situation where the SharePoint structure is clear, the sharing permissions are appropriate and the content is not too out of date, the outputs produced are both accurate and useful. But in an environment characterised by many years of uncontrolled accumulation of files, excessively broad sharing policies, and outdated documents, the outputs cannot be trusted, even though the budget remains unchanged. The amount of value it buys is not.

In the Copilot deployments that TrustedTech has looked at, the best predictor of poor Copilot performance isn't the product in question. Rather, it is a data environment which has never been set up in such a way that AI outputs can be made reliable. The situation is worsened by Cowork since it acts on the information it finds rather than simply displaying it. A preparation briefing based on six months' worth of clean CRM and email records is a valuable resource; one based on a chaotic SharePoint containing permission gaps and outdated files is a disadvantage.

The Copilot Readiness Assessment offered by TrustedTech checks whether an M365 environment is ready to support Cowork before any money has been spent on it; this is a more cost-effective initial move than finding out about the data quality issue after having spent three months' worth of credits on outputs which nobody could use.

Frequently Asked Questions

Q: Is the M365 Copilot license included with any Cowork credits?

The Microsoft Credits Guide makes it clear that the M365 Copilot USL provides no Cowork credit allowances; each Cowork credit used represents an additional cost beyond the $30 per user per month seat fee.

Q. Will the costs of working together decrease over time?

It is very likely that the costs will decrease over time due to three factors: as the technology develops, the models will become cheaper; Cowork will improve its ability to match the appropriate model to each task level; and both context retrieval and tool usage will become more efficient. The Cowork 1 model, which is being released shortly after general availability, is in fact designed to deal with ordinary task types at considerably lower credit consumption than Opus 4.8. Companies that are concerned about costs and who do not need to implement the solution right away can wait until Cowork 1 is available and then rework their budget estimate using the lower model cost.

Q. Can we set a limit on the amount that a single user spends each month?

Yes, MAC does allow per-user spending limits to be set through a group policy. If a user reaches their limit they can make a credit request from within Cowork for admin approval; however, the tasks which are currently in progress are not disrupted, the limit applying only to new tasks that are started after the limit has been reached.

Q. Are Copilot Credits eligible for MACC?

Yes. Both pay-as-you-go and the P3 prepaid plan are Microsoft Azure Consumption Commitment (MACC) eligible, per the Credits Guide. Credits can also be purchased through the CSP program for organizations managing their Microsoft licensing through a partner.

Q. What is the right time to switch from PayGo to prepaid?

After 60–90 days of PayGo usage, you have real consumption data to size a prepaid commitment. The P3 discount ranges from 5% to 20% depending on volume, but unused credits expire at the end of the term. Committing to a prepaid volume before understanding actual usage is a financial risk; the discount you fail to consume before expiry is a net loss against PayGo.

Cowork Related Articles

• Copilot Cowork FAQs: Everything IT Leaders Need to Know
• Which Copilot Cowork Skills Should IT Enable First?
• Is Copilot Cowork Worth It? An ROI Framework for IT Leaders