Table of Contents
- What Microsoft Copilot Usage-Based Billing Is
- The Two Timelines: November Frontier Rollout, December General Availability
- GPT-6 Astra and Claude Fable 5.1: The Two New Models
- Copilot Credits Explained: What They Cost
- The 4,000-Credit Default Limit and the 1 December Switch
- Who Gets Access — and Who Does Not
- What Copilot Credits Pay For (and What Stays in the Licence)
- How Admins Set Up Microsoft Copilot Usage-Based Billing
- What It Means for Your Bill: Three Worked Examples
- Reasoning Effort Control in Chat
- What Microsoft Has Not Said Yet
- What to Do Before 1 December: A Six-Step Checklist
- Frequently Asked Questions
- The Bottom Line
Microsoft Copilot usage-based billing is the biggest change to how businesses pay for Microsoft’s AI assistant since Copilot launched — and it arrives in two stages. From early November 2026, licensed users can pick two frontier models, OpenAI’s GPT-6 Astra and Anthropic’s Claude Fable 5.1, from the model selector inside Copilot Chat, Word, Excel and PowerPoint. Then, from 1 December 2026, new Microsoft 365 Copilot Business licences bought through a Cloud Solution Provider arrive with pay-as-you-go billing switched on by default, capped at 4,000 Copilot Credits per user per month — roughly $40 per user per month on top of the seat price.
For two years, Copilot was a predictable line in the budget: a number of seats multiplied by a flat monthly fee. That model is now splitting in two. Everyday AI — summaries, drafts, analysis and the automatic model picker — stays covered by the per-user licence. Advanced AI — long-running agent work, the new frontier models and a growing list of metered services — is billed by consumption through Copilot Credits. If nobody in your organisation owns that meter, the first invoice will be the first warning.
This guide explains exactly what Microsoft Copilot usage-based billing is, the two rollout dates you must not confuse, what GPT-6 Astra and Claude Fable 5.1 actually do inside Office, what Copilot Credits cost in US dollars, who gets access and who does not, and the six checks to run before 1 December — including the one-seat purchase that can switch the whole tenant over.

What Microsoft Copilot Usage-Based Billing Is
Microsoft Copilot usage-based billing — usually shortened to UBB — is a second way of paying for Copilot, on top of the per-user licence. Microsoft announced the framework in Message Center post MC1479276 on 25 September 2026, and the first user-facing result of it is Message Center post MC1519122, published 21 October 2026 and updated on 4 November 2026. That update is the document this article is built on, because it contains the two models, the two dates and the access rules.
The idea is simple, and it mirrors how cloud computing has been sold for years. A short email draft costs Microsoft very little to serve; a long-running agent task that plans, calls tools, reads documents and writes for an hour costs a great deal more. Under a flat seat price, the heavy user and the light user pay the same, and Microsoft absorbs the difference. Under Microsoft Copilot usage-based billing, the licence still covers the light, everyday work, while the heavy, advanced work is metered in Copilot Credits and billed by actual consumption.
That split is the single most important thing to understand about Microsoft Copilot usage-based billing. It is not a price rise on the existing licence, and it is not a replacement for it. It is a second layer. Nothing you do today stops working, and nobody is charged for Microsoft Copilot usage-based billing unless an administrator enables it and allocates credits. But the new frontier models — GPT-6 Astra and Claude Fable 5.1 — only exist in that second layer. If you want them inside Word, Excel, PowerPoint and Copilot Chat, you have to switch the meter on.
Three things make this change land harder than a normal Microsoft 365 update. First, the metered layer is visible to users even when it is switched off: people will see new models, banners about remaining credits and buttons to request more. Second, the default is changing for new buyers on 1 December 2026, so the decision arrives whether or not anyone schedules it. And third, one small purchase — a single new Copilot Business seat — can deploy Microsoft Copilot usage-based billing across every Microsoft 365 Copilot subscription in a tenant, according to Message Center post MC1476200. The rest of this guide works through each of those points in turn.
The Two Timelines: November Frontier Rollout, December General Availability
Microsoft Copilot usage-based billing has two different rollout schedules, and confusing them is the easiest mistake in this whole story. Microsoft’s own update on 4 November 2026 moved one of them — and several early news reports still show the old dates. Here are the two timelines, kept deliberately separate.
Frontier (Public) Rollout: Early to Mid-November 2026
The frontier rollout of Microsoft Copilot usage-based billing models began in early November 2026 and is expected to complete by mid-November 2026. This is the wave that puts GPT-6 Astra and Claude Fable 5.1 into the model selector for eligible tenants. “Frontier” here means the public preview wave: the models reach tenants whose administrators have already enabled Microsoft Copilot usage-based billing and allocated Copilot Credits, and they appear in Copilot Chat and in Copilot inside Word, Excel and PowerPoint.
If your tenant is in this wave and your administrator has done the setup, you may already see the two new models in the model selector this week. If your administrator has not enabled Microsoft Copilot usage-based billing, you will see nothing new at all — the models simply do not appear. That is by design, and it is worth repeating to anyone who asks why a colleague has GPT-6 Astra and they do not: access follows the billing configuration, not the calendar.
General Availability (Worldwide): Early to Mid-December 2026
General availability worldwide of Microsoft Copilot usage-based billing models begins in early December 2026 and is expected to complete by mid-December 2026. This is the date Microsoft moved. In the original announcement the worldwide rollout was scheduled for early to mid-November; the 4 November update pushed it to December. The frontier wave in November was not moved — only the worldwide general-availability window shifted by a month.
The same 1 December date carries a second, separate change for buyers. From 1 December 2026, new Microsoft 365 Copilot Business licences bought through a Cloud Solution Provider (CSP) arrive with Microsoft Copilot usage-based billing switched on by default, with a preset spending limit of 4,000 Copilot Credits per user per month. Microsoft first communicated 2 November as that date, then moved it to 1 December in its Partner Center announcements on 1 October 2026. Partner sandbox environments for testing the new purchase flow opened on 2 November instead.
So the calendar to pin on the wall is this: mid-November for the frontier models to finish reaching eligible tenants, and 1 December for the default-on switch for new Copilot Business purchases through CSP — with worldwide general availability of the models completing by mid-December. December is a rollout window, not a single-day switch, and tenants will see the models appear at different points inside it.
GPT-6 Astra and Claude Fable 5.1: The Two New Models
The headline of Microsoft Copilot usage-based billing is that two frontier models from Microsoft’s two biggest AI partners are now selectable inside the tools where office work already happens. Both become generally available as Microsoft Copilot usage-based billing models with this rollout, and both are subject to the eligibility requirements Microsoft previously communicated for them.
GPT-6 Astra: OpenAI’s Frontier Model Inside Office
GPT-6 Astra is OpenAI’s frontier model, and Microsoft Copilot usage-based billing makes it available in the model selector in Copilot Chat and in Copilot in Word, Excel and PowerPoint. Microsoft has published a “Available today: OpenAI GPT-6 Astra in Microsoft Copilot” announcement on its Community Hub, confirming the model is live for eligible users rather than a future promise.
For readers of this site, GPT-6 Astra sits in the same family as the GPT-6 models we have covered through the year — including GPT-6 Luna, the efficiency-focused model in the line-up. Astra is the frontier end of that family: the model you reach for when the task is hard, long or high-stakes, rather than the model you use for a quick rewrite. That is precisely why it is metered rather than bundled — it costs far more per request to serve than the models included in the licence, and Microsoft is passing that cost through as Copilot Credits instead of raising the seat price for everyone.
In practice, GPT-6 Astra inside Word means asking Copilot to work through a long contract, a research report or a messy draft with a stronger reasoning engine behind it. Inside Excel it means heavier analysis — multi-step reasoning over large tables, not just a formula suggestion. Inside Copilot Chat it means a model that can sustain a long, branching conversation without losing the thread. Whether that is worth the meter depends entirely on the task, which is why the reasoning-effort control covered later in this guide matters.
Claude Fable 5.1: Anthropic’s Model Joins the Selector
Claude Fable 5.1 is Anthropic’s frontier model, and it arrives in the same model selector, on the same terms, through Microsoft Copilot usage-based billing. Microsoft has likewise published an “Available today: Anthropic Claude Fable 5.1 in Microsoft Copilot” announcement, so this is a live, shipping model, not a roadmap item.
Fable 5.1 is the model users pick when the work is writing-heavy, nuanced or long-form — the kind of drafting, editing and analysis where Claude’s models have built their reputation. It is the same Anthropic family our site has followed all year, from Claude Message Threads, which cut waiting times on long-running agents, to the Claude agent features that now run inside Microsoft’s own tools. Having both an OpenAI frontier model and an Anthropic frontier model in one selector is the real product story here: for the first time, a Microsoft 365 user can choose between the two rival labs without leaving Word.
That choice is also a quiet competitive signal. Microsoft is no longer betting Copilot on a single model supplier. It is turning Copilot into a shelf of frontier models and letting Microsoft Copilot usage-based billing decide which ones earn their keep. If Astra and Fable 5.1 prove popular, expect more models to join the selector on the same metered terms — Microsoft has already said more usage-based agents and services will follow.

Copilot Credits Explained: What They Cost
Copilot Credits are the currency of Microsoft Copilot usage-based billing. Every metered action consumes credits, and the bill at the end of the month is the number of credits consumed multiplied by the price per credit. The unit is deliberately simple — one credit, one price — so that finance teams can budget it the way they budget cloud compute.
The Price: $0.01 Per Credit, With Bulk Discounts
Microsoft’s published list rate for Copilot Credits is $0.01 per credit on pay-as-you-go billing. That single figure makes the maths easy: 1,000 credits cost $10, 10,000 credits cost $100, and the default monthly limit of 4,000 credits per user works out to $40 per user per month at most.
Organisations that expect heavy use can pre-purchase credits instead of paying as they go. Microsoft offers a 5% discount on a one-year pre-purchase of 300,000 credits, rising to 20% off for 300 million credits. At the top end that is a substantial saving, but it is a commitment: pre-purchased credits are bought in advance, so the discount only pays off if the organisation actually uses them. For most small and mid-sized businesses, pay-as-you-go at $0.01 per credit is the sensible starting point, with a pre-purchase decision made after a month or two of real usage data.
One cost that Microsoft has not published is the credit rate per request for GPT-6 Astra and Claude Fable 5.1 themselves. The Message Center notice does not list how many credits a single Astra or Fable prompt consumes, and that rate may vary with the length of the request and the reasoning effort selected. Do not let anyone quote you a per-prompt price for these models yet — it does not exist in Microsoft’s published material. The honest way to budget is a pilot: give a small group a defined allowance, watch what they spend in a week, and scale from real numbers.
What Burns Credits Under Microsoft Copilot Usage-Based Billing
Understanding Microsoft Copilot usage-based billing means knowing exactly which actions draw on the meter. The metered list, confirmed across Microsoft’s licensing pages and the Message Center posts, is:
- Copilot Cowork — long-running, multi-step agent tasks that work on your behalf over minutes or hours.
- Work IQ APIs — programmatic access to Copilot’s work-graph intelligence.
- GitHub Copilot Harness — the agent harness for coding workflows.
- Copilot in SharePoint — agent experiences inside SharePoint.
- Frontier UBB models — GPT-6 Astra and Claude Fable 5.1 in Chat, Word, Excel and PowerPoint.
- Copilot Studio agents and Dynamics 365 agents, plus app hosting in the Copilot Managed Runtime, according to Microsoft’s licensing site.
What does not burn credits is just as important. Everyday Copilot stays inside the seat price: summaries, drafts, analysis, rewriting, meeting recaps and the automatic model picker are all still covered by the Microsoft 365 Copilot licence, which lists at $30 per user per month. Microsoft’s own summary of the split is that everyday AI stays at a fixed price per user while advanced AI runs on Copilot Credits. If your team only uses Copilot for the everyday tasks, Microsoft Copilot usage-based billing changes nothing on the invoice — provided nobody enables the metered features.
The 4,000-Credit Default Limit and the 1 December Switch
The number every finance lead should memorise is 4,000. From 1 December 2026, new Microsoft 365 Copilot Business licences bought through a Cloud Solution Provider arrive with Microsoft Copilot usage-based billing switched on and a default spending limit of 4,000 Copilot Credits per user per month. At $0.01 per credit, that is a maximum of about $40 per user per month in metered spend on top of the seat price — unless an administrator changes the limit.
To put that in perspective, Microsoft’s own example is that 100 Copilot Business users could consume up to 400,000 credits a month under the default limit — $4,000 a month of variable spend that did not exist in the budget before. For a 250-user business, the ceiling is 1,000,000 credits, or $10,000 a month. Those are ceilings, not forecasts: most users will spend far less, because most users will never touch a metered feature. But the ceiling is what the invoice can reach, and it is set by default, not by choice.
Who the Default Applies To
The default-on change is narrower than the headlines suggest, and the scope matters. It applies to new Microsoft 365 Copilot Business licences bought through CSP — standalone offers and bundles — in the commercial segment, in supported markets. Copilot Business is the tier for organisations with up to 300 users. The enterprise Microsoft 365 Copilot SKU is explicitly out of scope for the automatic set-up, and education, non-profit and Government Community Cloud customers are excluded too.
There is also a geographic limit at launch. The default will initially not be available in Australia, Belgium, Brazil, France, Germany, India, Italy, Korea, the Netherlands, Poland and Spain. Microsoft says it will share more as availability expands, and it has not published the full list of supported markets — so if your organisation buys licences in several countries, check each one rather than assuming the default applies everywhere.
The One-Seat Trap: A Single Purchase Can Switch the Whole Tenant
Here is the detail that catches organisations out. For existing customers, buying a single new Copilot Business seat deploys Microsoft Copilot usage-based billing across all Microsoft 365 Copilot subscriptions in the tenant, according to Message Center post MC1476200. The new licences also arrive with the Azure subscription setup that Microsoft Copilot usage-based billing needs, so there is no extra provisioning step to slow it down.
That means the trigger is not a big migration or a board decision. It is any planned seat purchase — a new hire, a team expansion, a licence true-up — landing after 1 December in a supported market through CSP. Consumption charges then flow through the partner’s invoice, and the transacting partner is responsible for all charges, including unexpected consumption. The spending policy is therefore a commercial conversation with your partner, not just an IT setting, and it needs to happen before the purchase, not after the first invoice.
Existing subscriptions keep their current terms, so nobody’s existing seats are silently repriced. But the moment a new seat is added under the new rules, the tenant has Microsoft Copilot usage-based billing — and the metered features become visible to users. The safe move is to decide the spending limit before any December purchase, even if the decision is to set the limit to zero.
The Same Date Brings Standalone Copilot Offers
1 December 2026 carries one more change that is easy to miss. From that date, new Microsoft 365 Copilot and Microsoft 365 Copilot Business offers can be purchased without a qualifying Microsoft 365 base subscription. The price stays the same, and the new offers replace the add-on offers for new purchases — the old add-on offers reach end of sale in CSP on the same day.
For a small business that wants Copilot without buying the full Microsoft 365 suite first, that is genuinely good news: it lowers the entry barrier. But it also means a new class of buyer — one with no existing Microsoft 365 estate and no licensing specialist — will meet Microsoft Copilot usage-based billing on day one, with the meter already on. If that buyer is you, read the checklist at the end of this guide before you click purchase.

Who Gets Access — and Who Does Not
Access to Microsoft Copilot usage-based billing models is gated three times over, and all three gates have to open. A user needs:
- A Microsoft 365 Copilot user subscription licence (USL) — the $30-per-user-per-month seat.
- Microsoft Copilot usage-based billing enabled for that user by a tenant administrator, through a spending policy.
- Copilot Credits allocated to that user, within the spending limit the administrator sets.
Miss any one of the three and the models do not appear. Microsoft states it plainly in MC1519122: users will not have access to UBB models unless UBB is enabled. There is no self-service switch for an individual user to turn the meter on for themselves — that is an administrator decision, made in the Agent 365 section of the Microsoft 365 admin centre, where spending policies are created and managed.
The gating works differently by tenant type, and the difference creates two opposite risks. In enterprise tenants, the metered features are visible to users but stay inactive until an administrator creates a spending policy — so the risk is confusion, not spend: people see new models and request buttons, ask for access, and the requests pile up with no owner. In Copilot Business tenants buying through CSP, the risk is the reverse after 1 December: the meter is on by default, so the risk is spend, not confusion. Both risks have the same fix — a named owner, a written policy and a communicated limit — applied before the rollout reaches your users.
There is one more access detail worth knowing. Microsoft says it plans to roll out the ability for users to request credits from their tenant administrators directly in the interface, after the initial rollout — but it has given no date for that feature. Until it arrives, credit requests are a manual, out-of-band process: an email, a ticket, a conversation. Name the person who approves them now, or the first request will sit unanswered.
What Copilot Credits Pay For (and What Stays in the Licence)
The cleanest way to explain Microsoft Copilot usage-based billing to a non-technical colleague is a two-column table. Everything in the left column is covered by the seat you already pay for; everything in the right column draws on Copilot Credits.
| Covered by the Copilot licence (no credits) | Metered — paid with Copilot Credits |
|---|---|
| Summaries, drafts and rewriting in Word, Outlook and Teams | GPT-6 Astra in Chat, Word, Excel and PowerPoint |
| Data analysis and formula help in Excel | Claude Fable 5.1 in Chat, Word, Excel and PowerPoint |
| Meeting recaps and follow-ups | Copilot Cowork long-running agent tasks |
| The automatic model picker | Work IQ APIs |
| Everyday Copilot Chat | GitHub Copilot Harness |
| Standard Copilot in SharePoint | Agent experiences in SharePoint that require credits |
| — | Copilot Studio agents, Dynamics 365 agents and app hosting in the Copilot Managed Runtime |
Two practical consequences follow. First, the licence still covers the work most people do most days, so the meter should not be feared — it should be managed. A team that never opens a frontier model and never runs a Cowork task will see a zero on the usage line. Second, the metered column is where the genuinely new capability lives. GPT-6 Astra, Claude Fable 5.1 and Cowork are not slightly better versions of the everyday tools; they are a different class of work — longer, deeper and more autonomous. The billing model is Microsoft’s way of saying that class of work has a different cost, and the organisation should decide deliberately who gets it.
One more migration note belongs here. AI Builder capacity add-ons were retired from sale on 1 November 2026, and Microsoft tells partners to plan for Copilot Credits once existing AI Builder credits run out. If your organisation has Power Platform flows or apps that use AI Builder features, those belong in the same budget conversation as Copilot Credits — they are heading for the same meter.
How Admins Set Up Microsoft Copilot Usage-Based Billing
Setting up Microsoft Copilot usage-based billing is an administrator’s job, done in the Microsoft 365 admin centre — specifically in the Agent 365 section, where spending policies live. The steps below describe the intended flow as Microsoft has documented it; the exact labels in the admin centre may vary slightly as the rollout progresses.
First, the administrator creates a spending policy. The policy names who gets Microsoft Copilot usage-based billing, how many credits each person or group may use, and which models and effort levels they may pick. Microsoft lets administrators assign models and effort levels by group, starting with Cowork — so a data team can get GPT-6 Astra while the wider company stays on the included models. That group-level control is the most powerful cost lever in the whole system, and it is worth using from day one rather than giving every licensed user the same allowance.
Second, the administrator sets the spending limit. The default is 4,000 credits per user per month for new CSP Copilot Business purchases, but it is adjustable — including down to zero. Setting the limit to zero turns Microsoft Copilot usage-based billing off for that group, which is the cleanest way to say “not yet” without removing the licence. A practical starting point, recommended by several Microsoft partners, is a low or zero limit for most users and a defined allowance for a small pilot group with a clear use case.
Third, the administrator connects the billing. New CSP purchases from 1 December include the Azure subscription setup that Microsoft Copilot usage-based billing needs, so for new buyers this step is handled at purchase. Direct-purchase customers have had Microsoft Copilot usage-based billing available since 19 October 2026, and enterprise tenants manage it through their existing Azure billing relationship. Consumption charges for CSP customers flow through the partner’s invoice.
Finally, the administrator watches the consumption dashboard. Microsoft has built the dashboard to link credit use to completed work, not just to raw consumption — so the monthly review should ask what the credits produced, not only how many were spent. Two or three measurable scenarios, such as RFP responses, contract reviews or meeting follow-ups, give the review something concrete to judge. If the numbers do not justify the spend, the limit comes down. That review belongs in a weekly or monthly routine with a named owner — which is why the checklist at the end of this guide starts with naming that person.
If you are the administrator reading this, one warning from the partner community is worth hearing: the metered features are visible to all users by default. People will see the new models, the credit banners and the request buttons. A short internal note explaining what is included, what costs extra and how to ask for access will prevent a week of confused tickets — and it will stop the shadow workarounds, such as staff pasting company documents into free consumer chatbots, that appear when the official route looks blocked.

What It Means for Your Bill: Three Worked Examples
Because Microsoft has not published per-request credit rates for GPT-6 Astra and Claude Fable 5.1, nobody can give you an exact monthly figure today. What we can do is show what the ceilings look like, so the worst case is visible before it arrives.
| Organisation | Seats | Seat cost ($30/user/month) | Maximum metered spend at the 4,000-credit default | Maximum total |
|---|---|---|---|---|
| Small business, 10 users | 10 | $300/month | 40,000 credits = $400/month | $700/month |
| Mid-sized company, 100 users | 100 | $3,000/month | 400,000 credits = $4,000/month | $7,000/month |
| Large business, 250 users | 250 | $7,500/month | 1,000,000 credits = $10,000/month | $17,500/month |
Read the table as a ceiling, not a forecast. The 100-user company will not spend $4,000 a month on credits unless its users actually consume 400,000 of them — and most will not, because most users will never open a frontier model. Realistic first-month spend for a company that gives a 10-person pilot group a 4,000-credit allowance each is $400 at most, and in practice far less. The point of the table is different: it shows that the default limit, left untouched, allows the metered line to exceed the seat line. That is the sentence to take into the budget meeting.
The other number to carry into that meeting is the pre-purchase discount. A company that settles at 300,000 credits a month of steady usage would pay $3,000 a month pay-as-you-go, or $2,850 with the 5% one-year pre-purchase discount — a $150 monthly saving for a commitment it can already justify from usage data. At very large scale the 20% tier is worth real money, but only after the usage pattern is proven. Buy the pilot first; buy the discount second.
Reasoning Effort Control in Chat
Buried in the 4 November update to MC1519122 is a small feature with a large effect on cost: users will be able to set the reasoning effort level when using a UBB model in Chat. Reasoning effort is the dial that tells a frontier model how hard to think — low effort for a quick answer, high effort for a problem that deserves sustained reasoning.
Why does a cost guide care about a reasoning dial? Because effort and consumption move together. A high-effort request consumes more compute, and under Microsoft Copilot usage-based billing it will consume more credits, than the same question asked at low effort. The dial therefore becomes a cost control as well as a quality control: quick questions go on low effort, and the expensive, high-effort setting is reserved for the analysis, the contract and the report that justifies it.
Administrators can assign effort levels by group in the spending policy, starting with Cowork — so the sensible policy is to give most users the low and medium settings and reserve high effort for the roles that need it, such as analysts working with long documents. It is also the feature to teach in the first training session: the difference between a $0.05 question and a $2 question is often just the effort setting, and most users will never have had to think about that before.
What Microsoft Has Not Said Yet
A guide built on Message Center posts has to be honest about the gaps in those posts. Microsoft Copilot usage-based billing still has several open questions, and any article that fills them in with guesses is doing its readers a disservice. Here is what is genuinely unknown as of 10 November 2026.
Per-request credit rates for GPT-6 Astra and Claude Fable 5.1. The rollout notice does not list them. Until Microsoft publishes them, budget with a pilot, not a spreadsheet.
The full list of supported markets. Eleven countries are excluded from the initial default-on change, and Microsoft has said it will share more as availability expands — but it has not published the complete list of markets where the default applies, or when the excluded countries join.
When users will be able to request credits from admins in the interface. Microsoft says the feature comes after the initial rollout, with no date attached. Until then, credit requests are manual.
Whether the December general-availability window will move again. It already moved once, from November to December. The frontier wave in November was not moved, which is a good sign — but a rollout window is a plan, not a promise, and tenants should check the Message Center rather than assume the date is fixed.
Which models join the selector next. Microsoft has said more usage-based agents and services will follow, and the Copilot licensing pages already list Copilot Studio agents and Dynamics 365 agents as credit-consuming. The direction of travel is clear — more of Copilot becomes metered over time — but the timetable is not.

What to Do Before 1 December: A Six-Step Checklist
Whether you are an IT administrator, a finance lead or a small-business owner buying Copilot for the first time, the same six checks apply. Run them before 1 December, not after the first invoice.
1. Find out how and where you buy Copilot. Ask your Microsoft partner or licensing owner three questions: do we buy Copilot Business or enterprise Microsoft 365 Copilot, do we buy through CSP or directly, and are any new seats or upgrades planned before year end? A single planned seat purchase is enough to trigger the default for the whole tenant.
2. Decide the real spending limit, per group. The 4,000-credit default is a ceiling, not a recommendation. Start low or at zero for most users, and give a defined allowance to a small pilot group with a clear use case. Remember: setting the limit to zero turns Microsoft Copilot usage-based billing off.
3. Decide who may use the new frontier models. GPT-6 Astra and Claude Fable 5.1 will only appear for users with billing enabled and credits allocated. Decide which roles need them — for example, analysts working with long documents — and write down why. Everyone else stays on the models included in the licence.
4. Name an owner for credit requests. Without a named owner and simple approval criteria, requests either stall or get approved by default. One person should own spending policies, credit requests and the monthly review of cost against outcomes.
5. Connect credits to outcomes, not just usage. Pick two or three measurable scenarios — RFP responses, contract reviews, meeting follow-ups — and review cost against result monthly. If the numbers do not justify the spend, lower the limit.
6. Tell employees what changed. Usage-based features are visible to all users by default. A short internal note on what is included, what costs extra and how to ask for access prevents frustration — and it keeps staff from pasting company documents into free consumer chatbots when the official route looks blocked.
Frequently Asked Questions
What is Microsoft Copilot usage-based billing?
Microsoft Copilot usage-based billing is a pay-as-you-go layer on top of the Microsoft 365 Copilot licence. Everyday AI stays covered by the $30-per-user-per-month seat, while advanced AI — GPT-6 Astra, Claude Fable 5.1, Copilot Cowork, Work IQ APIs and GitHub Copilot Harness — is billed in Copilot Credits at $0.01 per credit. It is enabled by an administrator and only charges for what is actually used.
When do GPT-6 Astra and Claude Fable 5.1 arrive in Copilot?
The frontier rollout began in early November 2026 and is expected to complete by mid-November 2026. Worldwide general availability begins in early December 2026 and is expected to complete by mid-December 2026 — a month later than first announced. Both models appear in the model selector in Copilot Chat and in Copilot in Word, Excel and PowerPoint, but only for users whose administrator has enabled Microsoft Copilot usage-based billing and allocated credits.
How much do Copilot Credits cost?
Copilot Credits cost $0.01 each on pay-as-you-go billing. The default spending limit for new Copilot Business licences bought through CSP from 1 December 2026 is 4,000 credits per user per month — about $40 per user per month at most. One-year pre-purchases earn a 5% discount at 300,000 credits, rising to 20% at 300 million credits. Microsoft has not published the per-request credit rate for GPT-6 Astra or Claude Fable 5.1.
Do I have to pay for Microsoft Copilot usage-based billing if I already have Copilot?
No. Existing subscriptions keep their terms, and Microsoft Copilot usage-based billing only applies where an administrator enables it and allocates credits. However, buying a single new Copilot Business seat can deploy Microsoft Copilot usage-based billing across all Microsoft 365 Copilot subscriptions in the tenant, so check the spending policy before any new seat purchase after 1 December 2026.
What happens if I never enable Microsoft Copilot usage-based billing?
Nothing changes on your invoice, and everything included in your licence keeps working. You simply will not see GPT-6 Astra or Claude Fable 5.1 in the model selector, and you will not be able to use Copilot Cowork, Work IQ APIs or GitHub Copilot Harness. The metered features stay visible in the interface, so your users may ask about them — which is worth a short internal note.
How do I turn Microsoft Copilot usage-based billing off?
An administrator or the CSP partner can set the spending limit to zero, or simply not create a spending policy. Users then keep everything included in the licence but cannot use experiences that require Copilot Credits. In enterprise tenants the features are visible but inactive until a spending policy exists, so “off” is the default state until an administrator acts.
The Bottom Line
Microsoft Copilot usage-based billing is not a price rise — it is a split. The flat seat still buys the everyday AI that most people use most days, and the meter buys the frontier models and agent work that cost far more to serve. GPT-6 Astra and Claude Fable 5.1 inside Word, Excel, PowerPoint and Copilot Chat are a genuine step forward: two rival frontier models, selectable in the tools where work already happens, with a reasoning-effort dial to control the cost of each request.
But a meter nobody owns is a bill nobody budgeted for. The dates are fixed enough to act on — the frontier rollout finishing mid-November, worldwide general availability completing by mid-December, and the default-on switch for new Copilot Business purchases on 1 December — and the numbers are simple enough to plan around: $0.01 per credit, 4,000 credits as the default limit, about $40 per user per month at most. Set the limit deliberately, name the owner, run a pilot before you scale, and tell your users what changed. Do those four things before December, and Microsoft Copilot usage-based billing becomes what it is meant to be: a way to buy exactly as much advanced AI as the work justifies — and not a dollar more.
Sources: Microsoft 365 Message Center post MC1519122 (published 21 October 2026, updated 4 November 2026); Microsoft 365 Message Center post MC1479276 (25 September 2026); Microsoft Partner Center October 2026 announcements on Copilot Business usage-based billing; MWPro analysis of MC1519122 (4 November 2026); Encorp analysis of Copilot Credits (2 November 2026). All prices are Microsoft’s published US list prices; per-request credit rates for GPT-6 Astra and Claude Fable 5.1 were not published as of 10 November 2026.
