SAMexpert logo
SAMexpert
Contact
Copilot

Copilot Cowork licensing and pricing

print
share

Summary

Copilot Cowork needs a Microsoft 365 Copilot licence, then charges for the work it performs through Copilot Credits at $0.01 per credit. In our 100-user example, consumption adds $363,000 a year to $36,000 in seats. What a task costs, when pre-purchasing credits pays off, and why spending limits do not stop charges instantly.

Your Microsoft 365 Copilot subscription is only the first part of the cost of Copilot Cowork. For work or school use, you need an assigned Copilot licence and enabled usage-based billing. Cowork then charges for the work it performs through Copilot Credits.

The published United States (US) enterprise Copilot price is $30 per user per month, paid yearly, while pay-as-you-go Cowork consumption costs $0.01 per credit. Dollar amounts in this guide are in US dollars (USD). In SAMexpert's illustrative 100-user budget below, consumption adds $363,000 a year to $36,000 in Copilot subscriptions. Your actual bill depends on the work, models and usage, so the useful question is how to build and control that second budget.

Availability, model choices, administration and security support below reflect published guidance as of 11 October 2026, unless another date is given.

Key takeaways

1. What you are buying

Copilot Cowork is the part of Microsoft 365 Copilot designed to carry out complex, long-running work across tools and deliver completed results. Charles Lamanna, Microsoft's Executive Vice President for Copilot, Agents, and Platform, announced worldwide general availability on 16 June 2026. You give Cowork work that can involve several steps and sources, rather than only asking for an answer in a single conversation.

Characteristic

Copilot Chat

Copilot Cowork

Main purpose

Always-on artificial intelligence (AI) for drafting, summarising and answering questions

An agentic system that carries out work and produces completed results

Execution time

Seconds to minutes

Minutes to hours, with autonomous execution

Task complexity

Single-step, single-session work

Multi-step workflows across tasks and sources

Source: Microsoft's Cowork frequently asked questions (FAQ).

You can use Cowork to draft, send, reply to and forward Outlook emails; schedule meetings and manage your calendar; create Word, Excel, PowerPoint and Portable Document Format (PDF) documents; and post to Teams channels and chats. Its documented capabilities also include searching across your organisation, deep research that synthesises multiple sources, scheduled recurring prompts and image generation using ChatGPT Images 2.0. Presentations can use your company's PowerPoint templates.

Microsoft packages these capabilities as skills. The built-in list comprises Word, Excel, PowerPoint, PDF, Email, Scheduling, Calendar Management, Meetings, Daily Briefing, Enterprise Search, Communications, Deep Research, Adaptive Cards and App. The App skill requires the Frontier programme. Microsoft's current guidance describes Frontier as opt-in access to preview capabilities before general availability, with your administrator enabling participation for the users who will evaluate those features. You can also provide up to 50 custom skills through SKILL.md files in your OneDrive /Documents/Cowork/skills/ folder.

You can access Cowork in a web browser at https://copilot.cloud.microsoft, the Microsoft Copilot desktop app on Windows or macOS, and the Microsoft Copilot mobile app on iOS or Android. The FAQ permits education faculty and staff, but excludes students. Although Microsoft described the June launch as worldwide, the FAQ limits Cowork to Anthropic-supported regions and says Cowork is not an exception.

The Copilot subscription includes Copilot Chat; Copilot in Word, Excel, PowerPoint, Outlook and Teams; the Work IQ context engine; access to multiple advanced AI models; the pre-built Researcher and Analyst agents; and custom agents built with Agent Builder. Cowork adds consumption charges to that subscription.

Your subscription prerequisite

For organisational use, Microsoft's prerequisites require an active Microsoft 365 Copilot licence assigned to your account, Cowork availability and usage-based billing, with Frontier participation additionally required for the App skill. These requirements concern work or school accounts; Microsoft's overview separately identifies a personal-account preview.

The US enterprise price for Microsoft 365 Copilot is $30 per user per month, paid yearly on an annually renewing subscription. You also need a separate qualifying Microsoft 365 plan. Checkout prices can differ by currency, country and regional variant.

2. How Cowork consumption becomes a bill

Microsoft's definition is “Copilot Credits are a common currency for eligible Microsoft services with usage-based billing”. You buy the subscription entitlement, then pay credits for Cowork's usage. Each task's cost reflects model use, context retrieval, tool calls and runtime: the model's work, the information it retrieves, the tools it invokes and how long it runs.

You consume credits when Cowork generates model responses, calls tools and skills, creates images or performs browser tasks. Each activity can involve several of the billing inputs above.

Microsoft offers two payment routes. Pay As You Go, or PAYG, charges $0.01 per Copilot Credit. The October guide confirms that you make no upfront commitment and pay in arrears for actual monthly consumption through the Microsoft Copilot Studio pay-as-you-go Copilot Credit meter, which covers eligible experiences, including Cowork. The Copilot Credit Pre-Purchase Plan (P3) instead offers a volume discount for an annual upfront commitment.

A task category is a planning aid

Microsoft defines three task levels. A light task uses a small number of knowledge sources, limited reasoning and one or fewer outputs. A medium task draws on multiple sources, applies structured reasoning and generates two or more outputs. A heavy task aggregates broadly, uses deep reasoning and produces many outputs.

Task level

Microsoft's illustrative credits per task

PAYG equivalent at $0.01 per credit

Light

100–300

$1–$3

Medium

300–700

$3–$7

Heavy

More than 700

More than $7

Sources: June GA infographic, October guide, pages 4 and 7. Dollar equivalents are arithmetic. Microsoft calls the credit figures rough planning estimates; actual pricing varies with workflows and usage patterns. The open-ended heavy category supplies no maximum task price.

Microsoft’s Copilot Credit Estimator forecasts monthly Cowork credit usage using personas and usage-intensity bands.

Microsoft's persona graphic describes four patterns of work:

  1. The Corporate Knowledge Worker juggles shifting priorities, a heavy email and chat load, and documents, presentations and spreadsheets.

  2. The Customer-Facing Knowledge Worker balances external customers with internal teams, organising work around accounts and relationship history.

  3. The Technical Worker needs deep focus, produces code, designs or systems, and works mostly outside Office applications.

  4. The Manager / Senior Leader needs headlines, risks and recommendations, decides more than creates, and has a calendar filled with high-stakes meetings.

SAMexpert's directional budget for 100 users

The illustrative workload uses Microsoft's published framework and assumes 100 licensed users completing three to five tasks per working day over 22 working days per month. The central case is four tasks per day, or 88 per month, with a light, medium and heavy task ratio of 20:10:2. That gives monthly averages per user of 55 light, 27.5 medium and 5.5 heavy tasks. The assumed average consumption is 200 credits for a light task, 500 for a medium task and 1,000 for a heavy task. The usage and credit allocations are SAMexpert's illustrative assumptions within Microsoft's broad planning categories, rather than measured customer consumption or Microsoft calculator defaults.

Average monthly work per user

Assumed credits per task

Calculation

Monthly credits per user

55 light tasks

200

55 × 200

11,000

27.5 medium tasks

500

27.5 × 500

13,750

5.5 heavy tasks

1,000

5.5 × 1,000

5,500

Total

11,000 + 13,750 + 5,500

30,250

Source: SAMexpert assumptions and arithmetic using the Microsoft task framework and October guide planning method.

The central case produces a combined annual budget of $399,000 for 100 users, before the separate qualifying Microsoft 365 plan. Cowork consumption is 30,250 × 100 × 12 = 36.3 million credits, costing $363,000 at $0.01 per credit. The $30 Copilot seat adds $30 × 100 × 12 = $36,000. Under these assumptions, consumption costs about 10.1 times as much as the seats.

Varying the workload from three to five tasks per working day gives a combined annual range of $308,250–$489,750. This sensitivity calculation keeps the assumed task mix and 22 working days per month, using the same published credit rate and Copilot seat price.

Budget measure

Three tasks/day

Four tasks/day: central case

Five tasks/day

Tasks per user/month

66

88

110

Credits per user/month

22,687.5

30,250

37,812.5

Consumption per user/month, unrounded

$226.875

$302.50

$378.125

Consumption plus seat per user/month, rounded

$256.88

$332.50

$408.13

Annual consumption for 100 users

$272,250

$363,000

$453,750

Annual consumption plus seats for 100 users

$308,250

$399,000

$489,750

Sources: SAMexpert's stated workload assumptions and arithmetic, using the published credit rate and Copilot seat price.

The $36,000 annual seat component stays the same because each scenario assumes 100 licensed users. Monthly combined amounts are rounded to the nearest cent; annual totals use unrounded monthly amounts.

A higher average cost for heavy tasks can increase your Microsoft AI budget even when task volumes stay unchanged. Keeping the central case's workload and credit rate, with an average of 5.5 heavy tasks per user per month, but increasing their assumed average from 1,000 to 2,000 credits adds 5,500 credits per user per month. Across 100 users, that adds $66,000 a year. Microsoft's heavy-task category has no upper credit limit, so the 1,000-credit average in our example does not cap your costs. Your bill depends on actual consumption.

3. When pre-purchasing credits saves money

P3 is a one-year, pay-upfront commitment. You purchase Copilot Credit Commit Units, or CCCUs. One CCCU represents $1 and 100 credits before the applicable discount. Consumption draws automatically from your pool, and unused credits expire annually.

The plan covers Microsoft Copilot Studio, Dynamics 365 first-party agents and Copilot Chat, and Copilot P3 credits can also pay for Cowork. Higher volume tiers provide progressively larger discounts. Cancellations and exchanges are unsupported, and all purchases are final.

The October 2026 Copilot Credits Guide lists nine credit volumes and discounts. The upfront amounts below are calculated from those figures at $0.01 per credit, so you can compare the annual commitment with your expected usage.

P3 tier

Historical CCCUs

October 2026 credit volume

Discount against PAYG

Calculated upfront amount (USD)

1

3,000

300,000

5%

$2,850

2

15,000

1.5 million

6%

$14,100

3

30,000

3 million

7%

$27,900

4

150,000

15 million

8%

$138,000

5

300,000

30 million

10%

$270,000

6

750,000

75 million

12%

$660,000

7

1,500,000

150 million

14%

$1,290,000

8

2,250,000

225 million

17%

$1,867,500

9

3,000,000

300 million

20%

$2,400,000

Sources: October 2026 Copilot Credits Guide, pages 4–5, for credit volumes, discounts and the rate used to calculate upfront amounts. The guide does not quote dollar totals. Historical CCCUs and matching upfront prices come from Microsoft's October 2025 FinOps schedule, published on 27 October and subject to change.

Under the October guide's terms, you pay upfront for one year, unused credits expire annually, and you can buy more credits or use PAYG once your pool runs out. Both PAYG and P3 count towards your Microsoft Azure Consumption Commitment, or MACC, which is your agreed Azure spending commitment. Eligible expenditure reduces the amount of that commitment you still have to meet.

For our assumed 100-user workload, pre-purchasing the listed 15-million-credit tier saves $12,000 against the $363,000 all-PAYG cost. It costs $138,000 upfront. The central case's 36.3 million annual credits would exhaust that pool, leaving 21.3 million credits charged through PAYG at $213,000. Total consumption spend would be $351,000, with no unused credits at annual expiry.

The purchase costs more than PAYG if annual usage falls below 13.8 million credits, the amount that $138,000 buys at the published PAYG rate. Above that usage, including in our central case, it produces a saving, assuming no other discounts. The nominal 8% discount therefore only becomes an actual saving if enough eligible usage consumes the pool.

P3 plans renew automatically by default. You can change their scope or ownership, but cannot split or merge them. Pre-purchase discounts do not combine with other discounts; Microsoft says applying P3 independently where you have additional discounts may produce an effective discount below the listed percentage. It also states that your realised cost per credit depends on utilisation. Those conditions belong in your Copilot pricing and negotiation assessment alongside the nominal rate.

A separate Microsoft Agent Prepurchase Plan covers a broader set of services: Copilot Credits, Microsoft Foundry, Microsoft Fabric and GitHub. Each Agent Commit Unit, or ACU, pays for $1 of eligible usage at its retail price. Microsoft illustrates its lowest tier with 20,000 ACUs covering $20,000 of usage for an estimated $19,000 upfront payment, a saving of approximately 5%.

Reservations apply before prepurchase plans, so only usage exceeding the reserved quantity draws from a prepurchase pool. Copilot Credit P3 then applies before the broader Agent plan where both cover the same usage. The Agent pool pays for the remaining eligible usage; it does not pay for purchasing the other reservations or prepurchase plans.

4. Testing Microsoft's claim of lower costs than Claude Cowork

Microsoft's June launch blog says “testing showed that Copilot Cowork on average was 30-40% cheaper” than Claude Cowork using the Microsoft 365 connector. The claim comes from an internal test of 125 runs across 12 light, medium and heavy prompts, with Opus 4.8 on both sides. Microsoft's Copilot calculation used internal logs for model use, context retrieval, tool calls and runtime. The Claude calculation used public application programming interface (API) rates and connector usage. An API lets software request work from a service, such as sending a message to Claude; the work requested then contributes to usage charges.

Microsoft did not publish the run-level inputs, token quantities or connector consumption needed to reproduce the claimed percentage. The comparison concerns that internal test; the difference between your subscription bills will depend on usage, configuration, time and other factors.

Tokens are the pieces of text a model processes. Input tokens cover the material sent to it, while output tokens cover what it generates. Anthropic's pricing page lists legacy Opus 4.8 at $5 per million input tokens and $25 per million output tokens. Prompt caching allows repeated input to be reused. Writing that input to a five-minute cache costs $6.25 per million tokens; reading the cached input for a later request costs $0.50 per million.

Anthropic's Pro and Team subscriptions include usage within plan limits, while Enterprise combines a seat charge with usage billed at API rates. The following US prices exclude applicable taxes.

Anthropic subscription

US price (USD)

Billing qualification

Pro

$20/month, or $200 paid upfront annually

Anthropic displays the annual option as approximately $17/month

Team Standard

$20/seat/month billed annually, or $25 billed monthly

Minimum two members

Enterprise

$20/seat/month

Billed annually, plus usage at API rates

Sources: Anthropic pricing, Pro plan pricing and Team plan pricing. Paid plans can purchase extra usage at API rates. Regional prices and tax treatment vary.

As of 11 October 2026, Anthropic describes Cowork and chat merging into one Claude, rolling out to Pro and Max first with more plans to follow, while its Team feature list still names Claude Cowork. Microsoft's comparison used the June configurations.

5. Model choice affects cost and deployment

You can influence consumption through model selection and reasoning effort. Microsoft also announced Cowork 1 in June, a fine-tuned model post-trained for Cowork tasks, as “designed to handle everyday Copilot tasks at a substantially lower cost” and “a secure, fine-tuned model releasing in the coming weeks”. Its release remains unconfirmed by the current model list, leaving that promised lower-cost option uncertain for your budget.

The model-selection page lists GPT 5.5, GPT 5.6 Sol, GPT 5.6 Terra, GPT 6 Astra, Opus 5, Claude Sonnet 5 and Claude Fable 5.1. GPT 5.5 is marked Frontier and is hosted on Azure AI Foundry. The GPT 5.6 and GPT 6 entries identify OpenAI as a subprocessor without specifying the same hosting arrangement. In that role, OpenAI operates the models under Microsoft's oversight; the designation identifies who processes the work, rather than a hosting location. The FAQ separately mentions Claude Fable 5 (Preview), off by default and requiring administrator enablement, and warns that some models require data retention.

For GPT 5.5, Microsoft's June release description specifies Azure hosting, enterprise security, compliance certifications and data-residency controls including the EU Data Boundary. Those assurances apply specifically to GPT 5.5.

Image generation uses ChatGPT Images 2.0.

Automatic selection and reasoning effort

Auto chooses among the models your organisation enables. On the web, Windows and macOS, reasoning choices include Light, Medium, High, Extra High and Max. Medium is the default, although the default can vary by model. Higher effort takes longer and consumes more credits, so your costs can rise even if your users submit the same number of tasks.

The credit averages in the 100-user budget are SAMexpert assumptions and do not specify a model or reasoning effort. Changing one of them, the heavy-task average from 1,000 to 2,000 credits, adds $66,000 a year for 100 users.

Microsoft documentation disagrees on model controls

Learn governance and access guidance say disabling Anthropic changes the available models while leaving Cowork accessible through permitted non-Anthropic models. The October guide, however, requires Anthropic to be enabled in your tenant, the Microsoft 365 environment managed for your organisation, as a Cowork prerequisite.

There is a separate scope conflict. Governance documentation and Microsoft's 25 September announcement describe user/group model-family controls. The access and model-selection pages still say the setting is tenant-wide and cannot enable a subset of users. The documentation therefore leaves unclear whether model permissions can differ between groups of users.

Cowork 1 and Microsoft's lower-cost models

Directions on Microsoft reported on 16 June that Microsoft officials described the planned Cowork 1 as “Azure-hosted”. As of 11 October 2026, Cowork 1's hosting and release remained unconfirmed in Microsoft's launch announcement and model guidance. The documented Azure-hosting arrangement for GPT 5.5 applies to that model.

Directions also interpreted Microsoft's strategy as positioning its own models as the lower-cost option across the Copilot family, discussing Cowork 1 alongside MAI-Code-1-Flash for GitHub. Microsoft's 2 June GitHub announcement described MAI-Code-1-Flash as a purpose-built small coding model. On 29 July, Microsoft separately reported 10% lower median token usage for MAI-Code-1-Flash in GitHub Copilot, and comparable quality to GPT-5.6 for the most common Excel tasks at significantly lower cost. Microsoft's figures concern those GitHub and Excel models. Directions' June interpretation is broader than those model-specific figures; the cited announcements do not describe a lower-cost Microsoft model for every Copilot product or specify those models' current availability.

6. Access, spending policies and the limits of budget controls

Your administrator enables usage-based billing in the Microsoft 365 admin centre and configures a spending policy that selects Cowork. The old Frontier or preview Cowork agent entry no longer controls access. If Cowork is discoverable but usage-based billing is not enabled, users can request access in the app for administrators to review and approve.

Discovery and permission are separate controls. A spending policy selecting Cowork grants access to every user within its scope even if its limit is only one credit. A policy can also grant access when discovery is off. If several policies select Cowork, precedence goes first to the highest per-user limit, then the largest overall policy limit, then the most recently created policy. Your effective spending permission can therefore depend on overlapping assignments.

Users can start additional tasks after reaching a credit limit because enforcement is asynchronous and takes effect after a delay. A displayed limit therefore does not promise an instantaneous stop at that exact amount. The Cost Management dashboard lets your administrators allocate credits centrally, apply policy-based access and limits, work across prepaid and PAYG models, monitor consumption and cost drivers, and identify policies or users approaching their limits.

At June launch, Microsoft said Cowork was off by default. It announced tenant, group and user spending limits; configurable usage alerts with administrator-defined thresholds and recipients; and the ability for users to request additional credits inside Cowork. Reporting was available at tenant, group and user levels, with breakdowns by user, group and feature. The announcement also stated: “Users see what each task costs as they run it”.

Current guidance gives you the free /cost command for approximate session consumption, monthly usage and remaining limits. It cannot price a task before execution, and its output is unsuitable as an invoice or chargeback record.

User limits reset monthly at 00:00 Coordinated Universal Time (UTC), unused monthly allowance does not carry forward, and an exhausted shared group allowance can prevent further use even when an individual balance appears available. Keep those monthly policy allowances distinct from the annual P3 commitment pool and its expiry.

Spending policies default to automatically covering future supported services and agents. Your administrator can turn off Auto-apply new services. The current billing overview includes Cowork, advanced work in SharePoint and OneDrive, Work IQ APIs, Copilot Managed Runtime and Teams Phone Agent, so the policy's service coverage is part of understanding what may consume your allowance.

Each cost signal in this guide arrives at a different point: task ranges before you start, /cost during a session, administrator reports afterwards. None of them gives a price for a task in advance.

Automated tasks and permissions

Cowork supports scheduled prompts and event-driven work triggered by matching emails or Teams messages. These tasks run as their creator. Consequential actions require approval by default, but users can pre-authorise actions. Rate limits, loop protection, auditing and Microsoft Purview policies apply.

Cowork inherits your signed-in Microsoft 365 permissions and can access only files and emails that you can already access. The FAQ limits file access and editing to OneDrive and SharePoint, excluding locally stored files. It says Cowork cannot delete files or folders in those services, and attached files must be less than 200 megabytes (MB). The FAQ also states a blanket inability to read encrypted files, even when you have access; the more detailed Purview documentation conflicts with that statement, as explained below.

7. Security, encrypted files and where work runs

Your governance assessment needs to establish where your team can inspect Cowork activity and whether its documented protections cover the files you intend to use. The current support matrix and the conflicting encrypted-file guidance set the boundaries for that assessment.

The current dedicated Purview support matrix includes Data Loss Prevention (DLP), auditing, Data Security Posture Management (DSPM) and DSPM for AI, sensitivity labels, encryption without sensitivity labels, Insider Risk Management, Communication Compliance, eDiscovery and Data Lifecycle Management. It marks Data classification and Compliance Manager unsupported. Cowork activity appears in activity explorer's AI activities tab, but does not appear in the Apps and agents dashboard or AI observability page. Those differences determine where your team can inspect the activity.

For new Word, PowerPoint and Outlook content, Purview documents sensitivity-label inheritance. The generated item automatically receives the source item's label and its protection settings, with the highest-priority label used when several sources are involved. Purview does not make the same inheritance statement for Excel.

Its encryption guidance requires VIEW permission to view the material and EXTRACT permission to copy or extract its content before the AI can return that data. Being able to open a protected document alone therefore does not ensure its text can be used in an answer. Purview also supports content encrypted using Customer Key or Bring Your Own Key (BYOK), arrangements involving your organisation's encryption keys. Those supported key arrangements do not remove the stated access-permission requirements.

The Cowork FAQ says encrypted files cannot be read, directly contradicting Purview's description of supported encrypted content. If your planned work depends on protected files, that conflict leaves their usability in Cowork unresolved, even when you hold the permissions Purview specifies.

Microsoft's launch description of Purview integration includes audit records for prompts, responses and generated artefacts, Insider Risk Management for user risk associated with Cowork interactions, and sensitivity-label display. The product release blog says: “No new stack to learn. No separate data security posture to manage. Admins govern Cowork the same way they govern Copilot today”. You still need to account for the specific support limits above, including the missing dashboards and unresolved encrypted-file support.

Cloud work and local browser work behave differently

Governance guidance says Cowork follows the same data-residency model as Microsoft 365 Copilot. It documents file processing in a temporary, isolated environment inside the Microsoft 365 service boundary, removed when the task finishes. Created files are saved to your OneDrive and SharePoint workspace, progress made while disconnected is preserved, and you can download output files to your device.

Local browser automation is generally available and requires Microsoft Edge installed and signed in. Cowork uses Edge on your device, your existing sign-ins and your organisation's existing web policies. Browser tasks inherit the conditional-access, DLP and tenant browsing policies already applied to you, and each task is recorded in the unified audit log.

If your device sleeps, a local browser task pauses. To resume, reopen the same Cowork conversation and send a follow-up. Cowork checks that Edge is available again and either continues or asks for confirmation.

8. Plugins and sales integrations have their own prerequisites

Plugin availability, administrator deployment and subscription prerequisites each affect what your users can do. Even with Cowork enabled, you may need an additional licence for a particular plugin, as the Sales comparison below explains.

You can build and deploy custom plugins, and bring existing Claude Cowork plugins into Copilot Cowork. Administrators can deploy plugins throughout the organisation using the Microsoft 365 admin centre, restrict which plugins are visible in the App Store, and deploy plugins that users cannot remove.

Microsoft Fabric IQ integration enables Cowork to access and work with Power BI data. The release blog identifies it as generally available, and the 25 September announcement additionally states that Fabric IQ is generally available in Chat and Cowork. Dynamics 365 integration at Cowork's June launch covered Sales, Customer Service and enterprise resource planning applications.

The Sales plugin and Sales for Cowork

The full Sales plugin requires Sales Enterprise or Sales Premium, an appropriate Microsoft Copilot access licence and the installed Sales plugin. The separate Sales for Cowork plugin provides skills that work without a customer relationship management (CRM) licence, while still requiring an appropriate Copilot licence. Cowork consumption is additional to the fixed licence and varies with prompts, duration and model.

For the full Sales plugin, your administrator must ensure that the Sales app is installed and available and enable the Dataverse Model Context Protocol (MCP) server for the environment. The Sales guidance also requires confirmation before consequential actions such as updating records, sending emails or creating tasks.

Microsoft’s skill lists show the practical distinction between the two plugins.

Plugin coverage

Listed skills

Both Sales and Sales for Cowork

Business case composer; Competitive strategy; Deal catch-up; Meeting preparation; Post-meeting follow-up

Additional skills in Sales

CRM data hygiene; Customer engagement strategy; Deal desk; Draft outreach; Executive deal review; Forecast rollup; Reconcile my opportunity; Sales Research Agent; Segment analysis; Stage advance readiness; Team activity coaching; Team pipeline review

Listed only in Sales for Cowork

Pipeline prioritizer

Source: Microsoft's Sales skills table.

9. Building an app and running it create separate charges

Microsoft's September app-building announcement introduced the Cowork /app skill through Frontier. Current Copilot Managed Runtime guidance retains preview status. If you create an app in Cowork, the build follows Frontier onboarding and your Cowork spending policy, with charges per user. In Copilot Studio, build activity is billed per environment, including authoring, testing and evaluation. Running the resulting app has separate per-user policies in the Microsoft 365 admin centre.

Power Apps Premium User Subscription Licences cover runtime within their existing request limits. Runtime beyond those limits consumes credits, as do AI-powered runtime services such as agents and Work IQ APIs. The Power Apps entitlement does not cover build charges. During preview, users who lack the required runtime credits receive a warning and are blocked after 20 app operations or five minutes, whichever comes first. Your build budget and your operating budget consequently need separate treatment.

App complexity

Illustrative credits to host, run and manage one app/month

Light

Approximately 10–50

Medium

Approximately 300–500

Heavy

Approximately 3,000–5,000

Source: October 2026 Copilot Credits Guide, pages 8 and 21. The runtime figures are rough consumption estimates, varying with task volume and complexity. Build consumption varies with model, complexity and authoring iteration. These estimates provide no fixed tariff for your app.

10. Work IQ APIs are another consumption boundary

Work IQ APIs reached general availability on 16 June 2026 with consumption-based billing through Copilot Credits. Microsoft states: “There is no separate Work IQ API subscription, SKU, or per-user license”. A stock keeping unit (SKU) identifies a separately listed product or purchasing option. Direct API charges apply when a developer builds an agent or app calling the APIs, or when you use a third-party agent that grounds its work in Microsoft 365 data through those APIs. That developer and third-party usage is distinct from your end-user Cowork interactions.

API

Credit consumption

Work IQ Chat API or Context API

Variable, depending on the scenario

Work IQ Tool API

0.1 Copilot Credits per API call

Sources: Work IQ announcement, October guide, page 13.

The announcement and October guide give light, medium and heavy scenario examples but publish no dollar prices or per-call dollar ranges for Chat and Context. Microsoft states: “These examples and prices are illustrative. Prices will vary based on the complexity of the scenario (inclusive of both Chat and Context scenarios)”.

Native Work IQ in Copilot chat, Office AI experiences and included agents such as Researcher, Facilitator and Analyst carries no incremental Work IQ charge; those experiences need not invoke Work IQ APIs. Work IQ APIs used in agents and AI solutions consume credits and are not entitlements in the Microsoft Copilot licence. Their prerequisites are tenant enablement and usage-based billing.

A temporary exception applies to the older Retrieval API and Chat API licensing model. The October guide retains that model for the time being, under which Microsoft Copilot-licensed users were not charged for their use.

Contractual restrictions on API use

The Microsoft IQ Product Terms, as of 11 October 2026, expressly license Work IQ API as a First-Party Consumption Service. You may not redistribute, resell or sublicense data obtained through the API, or use it to circumvent required Microsoft licences. The general API terms prohibit migration to non-Microsoft products while expressly preserving customer-authorised migration and export through Microsoft-supported capabilities. They also restrict performance testing without written permission, identifying or exploiting vulnerabilities or publicly disclosing them, misrepresenting licences and obscuring referral identifiers.

Third-party use is permitted solely in connection with a Microsoft AI Customer Solution, with you responsible for compliance and the necessary related licences. A Microsoft AI Customer Solution combines your application with a Microsoft AI Service and makes it available to your end users. Your application must add primary and significant functionality and must not primarily substitute for the Microsoft AI Service. Billing, licence management, prompt interfaces such as chat or workflows, and infrastructure services alone do not meet that functionality requirement. The same Microsoft IQ terms separately govern Web IQ approval and Azure AI Search Web Knowledge Source. Those provisions concern the services they name; they do not specify which services an individual Cowork interaction calls.

11. The forthcoming CSP billing default

From 1 December 2026, new Microsoft 365 Copilot Business purchases through the Cloud Solution Provider programme, or CSP, will include Azure setup and default PAYG billing. The change covers standalone and bundled purchases and eligible services including Cowork, Work IQ APIs and GitHub Copilot Harness. Microsoft's 1 October notice replaces the previously announced 2 November start date.

The default adjustable spending limit is 4,000 Copilot Credits per user per month, with charges based on actual usage. The 4,000 credits are a spending limit; Microsoft has not described them as a free included allowance. Initial exclusions are Australia, Belgium, Brazil, France, Germany, India, Italy, Korea, the Netherlands, Poland and Spain. Partner Center sandbox availability is scheduled for 2 November.

Microsoft's 10 September Partner Center notice also said the Cowork CSP Activation Incentive had launched on 1 September for eligible indirect resellers and direct bill partners driving adoption and usage, in addition to existing CSP incentives. It did not specify a customer discount or bonus amount.

The subscription-plus-consumption direction

Cowork follows the per-user-plus-consumption direction SAMexpert has already described. On Microsoft's 29 April 2026 earnings call, Satya Nadella said “And seat-based pricing is just entitlement to some consumption right” and “any per-user business of ours, whether it’s productivity, coding, security, will become a per-user and usage business”, while chief financial officer (CFO) Amy Hood described “a license business plus a consumption business”.

The GitHub Copilot usage-based transition began on 1 June 2026, while the announcement retained legacy annual Pro and Pro+ premium-request billing until plan expiry. Cowork and Work IQ APIs followed with their 16 June consumption launches. Those dated changes provide the context for treating your AI subscription and usage commitments together.

At Cowork's June launch, Microsoft said “more than half of the Fortune 500 is using Copilot Cowork”, and named Accenture, Avanade, Advance Local, Capital Group, Koch, LTM, Ooredoo Qatar and Zurich Insurance as early customers. On 29 July, it reported more than 30 million paid Microsoft 365 Copilot seats and thousands of customers already paying for and actively using usage-based Cowork. Nadella described the business model as per-seat-plus-consumption.

12. Frequently asked questions

Is Cowork included in my $30 Microsoft 365 Copilot subscription?

For organisational use, the Copilot licence is a prerequisite and Cowork adds usage-based charges through Copilot Credits. The published US enterprise $30 price is paid yearly and also requires a separate qualifying Microsoft 365 plan.

Can I know the exact cost before starting a task?

The /cost command reports approximate consumption and remaining limits, but cannot price your task before execution. Microsoft's light, medium and heavy ranges are planning estimates, with an open-ended heavy category.

Will a spending limit stop charges immediately?

Enforcement is asynchronous, so additional tasks can start after you reach a limit before the block takes effect. Your effective policy also depends on the precedence rules for overlapping policies.

Does a pre-purchase discount guarantee a lower annual bill?

Your realised saving depends on utilisation because you pay upfront and unused credits expire annually. The listed 15-million-credit tier costs $138,000 upfront and costs more than PAYG if annual usage falls below 13.8 million credits, assuming no other discounts. Our assumed 100-user central case consumes 36.3 million credits, exhausting the pool and adding $213,000 in PAYG charges. The resulting $351,000 consumption spend saves $12,000 against all-PAYG billing, with no unused credits. That comparison uses the published rate and tier with our explicit workload assumptions.

Can SAMexpert help assess our Cowork budget?

If you need to assess a proposed Cowork rollout, we can help you work through the seat commitment, usage assumptions, credit purchases and spending controls together.

Contact SAMexpert for independent advice on your Copilot licensing and consumption budget. We do not sell Microsoft licences or cloud services, so our advice focuses on your costs and commercial interests.

Table of contents
print
share

Read next

More articles
Share this article