Copilot Credits are now the single meter for agentic AI across the Microsoft estate. One cent each, pooled at the tenant, and they burn down your Azure commitment. This pillar prices the meter and maps the EA negotiation levers that keep it predictable.
Microsoft has named the meter. Copilot Credits are the single currency for agentic AI across Cowork, Copilot Studio, Dynamics 365 agents, and the Work IQ APIs, all on one pooled tenant balance at one cent per credit. They decrement your Azure commitment like any other Azure spend, which makes them easy to approve and easy to lose track of. This pillar prices the meter and sets the buyer side moves before your next renewal.
Copilot Credits are a usage based meter that sits alongside the per user Microsoft 365 Copilot subscription. The subscription still covers everyday Copilot Chat, the in app experiences in Word, Excel, PowerPoint, Outlook and Teams, and the built in agents. Credits are charged on top, for the agentic work that runs autonomously and at variable cost.
They are pooled at the tenant level, so all consumption draws on one shared balance that administrators manage centrally. Microsoft documents the meter, the one cent rate, and the purchase options in its own Copilot Credits overview. The currency itself is not new packaging on the old model. It replaced the prior message based meter for agents in September 2025.
One balance covers a widening set of surfaces. That breadth is the point: Microsoft wants one meter for every agent you run.
For a buyer, the unified pool is a double edged thing. It simplifies the invoice to one line. It also concentrates every AI workload onto a single meter that the vendor controls the pricing of.
A credit looks like one cent, one unit, one invoice. What it hides is the bundle. Each credit packages the model with the runtime that orchestrates the agent, the context it retrieves from your mail and files, and the tool actions it takes.
That is why a credit bills by how hard the task was, not by what you sent and received. You cannot reverse a credit into a clean token rate, which is exactly what makes the meter hard to benchmark against a direct model price. We return to that comparison in the buy or build section.
There are three purchase routes, and the choice is not whether credits touch your commitment. All three decrement your MACC. The choice is whether you stay flexible or commit up front for a discount.
The three ways to buy Copilot Credits, and what to watch
| Option | How it works | MACC | What to watch |
|---|---|---|---|
| Pay as you go | One cent per credit, billed monthly in arrears for exactly what you used | Yes | Cost is fully variable and only visible after the fact |
| Capacity Packs | 200 dollars per tenant per month for 25,000 credits, reset monthly | Counts toward spend | Unused credits reset each month, so steady volume only |
| Pre-Purchase Plan | Annual pool at a 5 to 20 percent tiered discount, paid up front | Yes | Discount caps at 20 percent and unused credits expire at term end |
The published rate and the three routes are set out in Microsoft's Copilot Studio billing and licensing documentation and the Copilot Studio pricing page. Read both before you accept a sizing from your account team.
The Pre-Purchase Plan, P3, runs a tiered discount that climbs with commitment size. The detail matters, because the headline 20 percent is only available at the very top of the ladder.
Pre-Purchase Plan tiers and the effective per credit price
| Tier | Credits | List price | Discount | Effective per credit |
|---|---|---|---|---|
| 1 | 300,000 | $3,000 | 5% | $0.00950 |
| 3 | 3,000,000 | $30,000 | 7% | $0.00930 |
| 5 | 30,000,000 | $300,000 | 10% | $0.00900 |
| 7 | 150,000,000 | $1,500,000 | 14% | $0.00860 |
| 9 | 300,000,000 | $3,000,000 | 20% | $0.00800 |
The prepay floor is eight tenths of a cent
Effective price per credit at each Pre-Purchase tier. Even a three million dollar commitment never pays less than 0.8 cents.
Tier detail: Microsoft Copilot Credit Pre-Purchase Plan, 2026. Analysis: Redress Compliance.
Hold that floor against what direct model optimization can reach. The deepest prepay still pays four fifths of list. On a direct model contract, batch processing and prompt caching can cut the effective rate far below that, which is the heart of the buy or build decision. Microsoft introduced and documents the plan in its Pre-Purchase Plan announcement.
Microsoft prices by task complexity, not by tokens, and publishes illustrative ranges. Converting at one cent per credit gives the real dollar cost of the work. The spread between a light and a heavy task is more than seven to one.
Cowork task tiers in credits and dollars at list
| Intensity | What it does | Credits per task | Cost at one cent |
|---|---|---|---|
| Light | Draft a short update from your calendar and recent notes | 70 to 200 | $0.70 to $2.00 |
| Medium | Build a meeting brief from emails, CRM notes, and a deck | 400 to 600 | $4.00 to $6.00 |
| Heavy | Analyze a large data export and write a leadership summary | Over 1,500 | Over $15.00 |
The intensity maps to how much the agent reads and writes. A light task reads roughly a short stack of context and writes a paragraph. A heavy task can ingest hundreds of pages of context before it writes a full analysis. Input tokens are everything the model reads. Output tokens are what it writes back.
Convert credits to dollars on your real task mix, by persona. The arithmetic is simple and it is the single most useful number you can put in front of finance.
We model this per deployment for clients before provisioning. A worked model and the persona template live in the cost per task companion guide. Build the number before Microsoft builds it for you.
A credit looks like one cent. What it hides is the model, the runtime, the context, and the tools, bundled into one number that bills by how hard the task was.
White Paper · Microsoft
What Microsoft Copilot Cowork Really Costs
What a task really costs in dollars, the prepay floor, and the same work on Claude direct. Read it free.
This is where the meter meets the contract, and where the buyer side leverage actually sits. Both prepay and pay as you go decrement your Microsoft Azure Consumption Commitment, the MACC, exactly like other Azure spend. For an EA customer with a large unconsumed commitment, that is genuinely useful. It turns committed dollars you were going to forfeit into AI capability.
It is also the reason governance slips. Spend that draws against an already committed pool feels free at the point of use. It is not free. It is commitment you could have spent on something you chose deliberately, or renegotiated down.
Three things move when credits enter the EA conversation. Each is a lever, and each is covered in depth across the cluster.
The Copilot estate does not sit on its own. It sits inside the Microsoft EA and the Azure commitment you are already negotiating. Treat the credit meter as one line in that deal, not a separate purchase your account team frames in isolation.
Source: Redress Compliance advisory engagement file, 2024 to 2025, against Microsoft published pricing.
Same model, different contract. You can consume Claude through Microsoft, baked into Copilot and Cowork or as a model in Azure Foundry, or you can sign directly with Anthropic and consume the API. The trade is managed convenience on one side against economics and engineering on the other.
Put real dollars on it. Take a realistic deployment of 200 Cowork users, each running about 20 light, 10 medium, and 2 heavy tasks a month. That is roughly 27 million credits a year. Priced four ways, the gap is stark.
Same year of work, four ways to pay
200 users, about 27 million credits a year. Microsoft credit paths versus the same inference on Claude direct at Anthropic list rates.
Anthropic list rates, June 2026, via the Anthropic pricing page. Model: Redress Compliance.
The honest read is that the gap is not Microsoft overcharging for the model. It is the price of everything bundled around it: the runtime that orchestrates each agent, the retrieval that grounds it in your data, the tool execution, the identity, and the governance. On Claude direct you pay only the inference. To match what Cowork does, you build and run those layers yourself.
The real question is build versus buy, and it turns on volume and team. Capturing that quarter of a million means standing up and operating an agent platform.
The full comparison, including the two processor model where Claude runs through Foundry, is in the credits versus Claude direct guide and the direct versus managed analysis.
Variable per task billing without caps is how AI becomes the line item nobody approved. The governance is not complicated, but it has to be in place on day one, before the first agent runs.
The standard account team pitch is to prepay a large Pre-Purchase tier so you lock the discount and never run out. We disagree. In the renewals we benchmarked, the prepay discount tops out at 20 percent while unused credits expire at term end, so a pool sized on a vendor forecast routinely wrote off 10 to 30 percent of its value. Pay as you go combined with direct optimization beats that almost every time at low and moderate volume. The buyer side move is to model your real task mix in dollars, turn on spend caps, buy pay as you go, and let demonstrated consumption, not a forecast, justify any prepay. Commit to volume you have proven, never to volume the vendor projects.
The full control set, the admin center walkthrough, and the renewal clauses to ask for are in the EA negotiation and governance guide.
The per user Microsoft 365 Copilot license and the credit meter are two distinct lines, and conflating them is the most common budgeting error we see. The subscription is a prerequisite that unlocks the in app and Cowork experiences. The credits are the variable consumption charged when agents do autonomous work.
The subscription itself is not static, and the 2026 changes matter for renewal timing. The Copilot add on lists at 30 dollars per user per month and requires an eligible base license underneath it. The base is moving too.
The three cost layers on one Copilot user
| Layer | What it is | Shape | Buyer move |
|---|---|---|---|
| Base license | E3 or E5, the prerequisite under Copilot | Per user, rising July 2026 | Right size E3 versus E5 before adding Copilot |
| Copilot add on | The 30 dollar per user per month seat | Per user, volume discount time bound | Lock the add on discount inside the EA term |
| Copilot Credits | The agentic consumption meter | Variable, per task | Cap and govern, buy pay as you go first |
The renewal lesson is to negotiate the three layers together, not in sequence. An account team that books the add on and then frames credits as a separate purchase has split your leverage in two. We track the base increases and the add on window in the 2026 price change analysis, and the seat economics in the Microsoft 365 Copilot pillar.
The Work IQ APIs went generally available on June 16, 2026, and they bill from the same credit pool as everything else. That matters because Work IQ is the grounding layer: it lets agents and custom applications reason over your mail, meetings, documents, and chats while preserving existing permissions. Every grounded query draws credits.
Work IQ pricing has a fixed component for tool actions and a variable component for queries. The query tiers are smaller than Cowork tasks, but they multiply fast across a custom agent estate.
Work IQ API consumption in credits and dollars
| Call type | Intensity | Credits | Cost at one cent |
|---|---|---|---|
| Query | Light | 20 to 40 | $0.20 to $0.40 |
| Query | Medium | 30 to 75 | $0.30 to $0.75 |
| Query | Heavy | 50 to 150 | $0.50 to $1.50 |
| Tool action | Per call | 0.1 fixed | $0.001 per call |
Because it scales with automation, not with headcount. A single custom agent that runs thousands of grounded queries a day can quietly become a larger credit draw than your human Cowork users.
The Work IQ general availability terms and the credit denomination are documented in Microsoft's Work IQ APIs announcement. Treat any developer led adoption as a budget event, not a free feature.
There is no single right answer. The right path depends on your Azure commitment position, your task volume, and whether you already run an agent platform. The profiles below map the buyer to the move.
Match your profile to the path that fits
| Profile | The path that fits |
|---|---|
| Underconsuming MACC holders | Credits convert committed Azure spend into capability. Model the task mix, then buy pay as you go and let consumption prove the volume before any prepay. |
| High volume agentic teams | Build on Foundry or direct Claude. Pay token rates, apply batch and caching, and reserve credits only for turnkey Microsoft workflows not worth building. |
| Light or occasional users | Pay as you go credits. Do not prepay. Flexibility beats a 5 percent tier one discount that risks expiry. |
| Regulated and governance led | Credits give one invoice, Entra identity, and central spend controls. Weigh that against direct, and map the two processor model where Claude runs through Foundry. |
| Cost led, high volume | Credits are rarely cheapest per unit of inference. Model the effective rate on your mix before concentrating spend on the credit meter. |
Whichever profile fits, the discipline is the same. Decide per workload, keep the model and build choice open, and never let a credit prepay lock your AI strategy to one path. The strength of the Microsoft estate is that both Claude and GPT run on one commitment. Spend that flexibility deliberately, do not give it away for a single digit discount.
Cowork is the visible face of the meter, but it is not the whole draw. The same tenant pool funds the custom agents your teams build in Copilot Studio, the Dynamics 365 first party agents, and Copilot Chat for users who do not hold a paid seat. Every one of them spends the same credits.
That breadth is why a credit budget built only on human Cowork usage understates the real number. The agents you cannot see on a seat report can be the largest consumers on the invoice.
Custom agents scale with automation, not headcount, so they break the per user mental model. A single Copilot Studio agent wired into a busy process can outspend a department of human users.
Copilot Chat usage by users without a paid Microsoft 365 Copilot license is metered to the same credit pool rather than to a seat. That is useful, because it lets occasional users touch Copilot without a full subscription. It is also a quiet draw that no seat count will reveal, so it belongs in the forecast and under the same caps.
Walk a representative renewal. The numbers are illustrative, but the moves are the ones we run. The shape repeats across the engagements behind this pillar.
A 5,000 seat enterprise is 18 months into a three year EA with a sizable Azure commitment it is tracking behind on. The account team proposes a Tier 6 Copilot Credit prepay, 75 million credits for 660,000 dollars at 12 percent off, sized to a Copilot adoption forecast the vendor built.
The pitch is framed as a discount and as a way to put the underused commitment to work. Both halves contain a real point and a real trap.
The response is not to reject credits. It is to separate what is proven from what is projected, and to fold the meter into the existing agreement.
Vendor proposal versus buyer side counter
| Lever | Vendor proposal | Buyer side counter |
|---|---|---|
| Volume basis | 75M credits on a forecast | 30M credits, proven from the task mix |
| Purchase model | Tier 6 prepay, paid up front | Pay as you go against the existing MACC |
| Rate protection | Discount at signing only | Rate and discount held for the term |
| Governance | Set up after provisioning | Caps and attribution before day one |
The outcome is the capability without the forfeiture. The buyer recovers committed dollars that were heading for waste, keeps the flexibility to optimize, and avoids writing off a prepay pool sized to a number it did not build. The discount it gave up was never worth the lockup.
Copilot Credits are part of a broader industry shift from per seat software to per use metering, and the same discipline applies across all of it. Microsoft itself now meters multiple products this way, which is the strategic point: usage billing concentrates spend onto vendor controlled meters.
The buyer side lesson is consistent with the rest of this pillar and the wider Microsoft EA playbook. Treat every usage meter as a line to model, cap, and benchmark, never as a flat fee to wave through.
Use this sequence. It works whether you are evaluating Copilot or 90 days from an EA renewal.
Copilot Credits are Microsoft's metered currency for agentic AI, priced at one cent per credit and pooled at the tenant level. One shared balance covers Cowork tasks, Copilot Studio agents, Dynamics 365 first party agents, the Work IQ APIs, and Copilot Chat for unlicensed users. They are charged on top of the per user Microsoft 365 Copilot subscription, not instead of it.
A Copilot Credit lists at one cent on pay as you go. The Pre-Purchase Plan discounts that by 5 to 20 percent depending on commitment size, so the floor is eight tenths of a cent at the deepest tier. Capacity Packs sell 25,000 credits for 200 dollars per tenant per month, which is also one cent per credit.
Yes. Pay as you go credits, Capacity Packs, and the Pre-Purchase Plan all decrement your Microsoft Azure Consumption Commitment, the MACC, exactly like other Azure spend. That makes credits easy to approve against committed dollars and easy to lose track of without spend governance.
Microsoft prices by task complexity, not by tokens. A light Cowork task runs 70 to 200 credits, a medium task 400 to 600, and a heavy task over 1,500. At one cent per credit that is under two dollars for a light task and over fifteen dollars for a heavy one.
Rarely on a forecast. The Pre-Purchase Plan discount tops out at 20 percent and unused credits expire at the end of the one year term. Pay as you go plus direct optimization usually beats a prepay that overshoots. Commit only to volume you have already demonstrably consumed.
No, not per unit of inference. The same task often costs several times more through credits than on Claude direct at Anthropic token rates, because a credit bundles the model with the runtime, retrieval, tools, identity, and governance. The gap is the price of buying the managed agent rather than building it.
No. The per user Microsoft 365 Copilot license, listed at 30 dollars per user per month, is a prerequisite for the in app and Cowork experiences. Copilot Credits are a separate, variable consumption meter charged on top. Budget the seats and the per task credits as two distinct lines.
Turn on spend governance before you provision. Set spend policies, per persona thresholds, and caps in the admin center so variable per task billing cannot run unmonitored. Convert your real task mix to dollars first, then buy pay as you go and let consumption prove the volume before any prepay.
The task mix model in dollars, the credit to dollar conversion across light, medium, and heavy work, the build versus buy math against Claude direct, and the governance controls to set before you provision.
Used across more than five hundred enterprise engagements. Independent. Buyer side. Built for procurement leaders sizing Copilot inside the next EA renewal.
500+ enterprise clients. 11 vendor practices. We will model your task mix in dollars, weigh direct Claude against buying through Microsoft, and set the governance and commitment terms before Microsoft sizes them for you.
One email a month on Copilot Credits, EA renewal levers, Azure commitment, and audit defense. Buyer side only.
