Agentforce vs Copilot Studio Pricing: Cost per Action, Credits and Seats Compared

The short answer: a single agent action costs $0.10 on Agentforce and $0.04 on Copilot Studio, but Microsoft itemizes more of what happens inside one turn, so per-job the two land closer than the headline rates suggest. The decision usually comes down to three structural differences instead: Microsoft zero-rates employee use for anyone who already holds a Microsoft 365 Copilot license, Salesforce lets prepaid credits live for the whole contract term while Microsoft expires them every month, and Microsoft switches your custom agents off at 125% of prepaid capacity while Salesforce publishes no equivalent cutoff.

Both vendors have now settled on the same billing philosophy. Neither sells an AI agent by the seat. You buy a pool of prepaid credits, agents draw from it as they work, and the seat license mostly decides whether a given person's usage draws down that pool at all. Everything below is US list pricing, read off Salesforce's own pricing page with the currency selector set to US Dollar and off Microsoft's published Copilot Credit rate card, checked in August 2026.

The two rate cards side by side

Start with the credit itself, because both vendors quote everything in their own unit and the units are not the same size.

Read the "one agent action" row and you would conclude Microsoft is 60% cheaper. Read the next two rows and the picture changes, because Salesforce folds a lot into the word action. Salesforce defines it as "a specific function that an AI agent executes on the platform, such as updating a record, summarizing a complex case, answering a product inquiry, or executing a custom prompt or flow." One flat charge covers the retrieval, the reasoning and the write. Microsoft bills the generative answer, the grounded lookup and the action separately, so one turn of a Copilot Studio agent frequently draws from three meters at once.

Is Agentforce cheaper than Copilot Studio?

It depends entirely on how many billable events one job produces, and for the most common shape of work the two come out within pennies of each other. Modelling a single service interaction where the agent looks up an order, summarizes the case and updates a record gives you this, with assumptions stated rather than hidden.

These are our models, not vendor figures, and the event counts are the assumption you should argue with rather than the prices. On these assumptions Microsoft comes out roughly half the cost, and that holds up as long as your agents mostly reason and retrieve. It stops holding the moment an agent does a lot of small discrete platform operations, because Salesforce charges one flat $0.10 for a whole compound job while Microsoft charges $0.04 for each of the five steps inside it.

The honest conclusion is that unit price is the wrong thing to negotiate. Count billable events per job for your top three use cases, price both rate cards against that count, and treat the result as your comparison. Anyone quoting you a per-credit price without a job model is selling you an unknowable number.

What about the $2 conversation option?

Salesforce also sells a flat Conversations meter at $2.00 per conversation for customer-facing agents, and it has one clean property: the break-even against Flex Credits is exactly 20 actions. Below 20 actions per conversation, credits are cheaper. Above it, the flat rate is. Most routine support interactions resolve in far fewer than 20 platform actions, so for the majority of deployments the flat rate is expensive certainty. Microsoft offers no equivalent, so if a predictable per-conversation number is what your finance team wants, only one of these two vendors can give it to you. The full breakdown of both Salesforce meters, including the seat options, is in our Agentforce pricing analysis.

Which is cheaper for employee-facing agents?

This is where the gap stops being a rounding difference and becomes a strategic one. Microsoft zero-rates employee-facing agent use for any user who already holds a Microsoft 365 Copilot license. Those interactions do not draw Copilot Credits at all. Salesforce has no equivalent zero-rating. Its employee options are a $5 per user per month license with metered usage that still consumes Flex Credits, or a $125 per user per month add-on that makes employee usage unmetered.

If your company already pays for Microsoft 365 Copilot, internal agents on Copilot Studio are close to free at the margin, and that is very hard for Salesforce to answer on price alone. What Salesforce answers with instead is proximity to the CRM record. An Agentforce agent updating an opportunity is operating inside the system of record with the permissions model already applied. A Copilot Studio agent reaching the same record is going through a connector. Whether that is worth $125 a seat is a real question with a real answer, and the answer differs by company.

Going the other direction, Agentforce 1 Editions start at $550 per user per month and include 2.5 million Flex Credits. Check the unit before you value that: Salesforce lists it as per org per year, not per user. That is $12,500 of credits and about 125,000 actions for the whole company for a year, so at 100 seats it is worth under 2% of the annual seat bill.

What happens to unused credits?

Both vendors expire them, but on clocks that differ by an order of magnitude. Salesforce says unused Flex Credits "do not roll over into subsequent subscription terms," which on a standard annual contract gives you twelve months to smooth out demand. Microsoft expires unused Copilot Credits monthly.

For a business with flat, predictable agent volume this barely matters. For a seasonal one it matters enormously. A retailer whose support volume triples in November either buys peak capacity every month of the year on Microsoft and writes off the other eleven months, or runs pay-as-you-go at roughly $0.01 a credit instead of $0.008 and pays a 25% premium for the flexibility. On Salesforce the same retailer buys an annual pool and simply uses more of it in Q4.

The overage behavior is asymmetric too. Microsoft publishes a hard consequence: at 125% of prepaid capacity, custom agents are disabled. That is a service outage triggered by a billing threshold, and it needs an owner and an alert before you go live. Salesforce publishes no equivalent cutoff on its pricing page, which is not the same as saying there is none, so get the overage behavior written into your order form rather than assumed.

Both vendors bill metered usage monthly in arrears, which means the invoice always arrives after the spend. Salesforce ships Digital Wallet free with supported products for near real-time consumption tracking and threshold alerts, and Microsoft exposes Copilot Credit consumption in the admin center. Turn the alerts on during the pilot, not after the first surprise, and have someone pull the consumption report and reconcile it against what you were billed each month, the same routine discipline any finance team applies when it matches transactions back to the ledger. Our breakdown of the Microsoft side, including the full credit rate card and the three runtimes that bill differently, is in the Copilot Credits pricing guide.

The marketplaces both vendors want you to buy from

Neither company expects you to build every agent yourself, and both renamed their marketplace in the last two years to say so. Salesforce turned AppExchange into AgentExchange, and as of August 2026 appexchange.salesforce.com literally carries the title "Salesforce AppExchange is now AgentExchange." Microsoft merged AppSource and Azure Marketplace into a single Microsoft Marketplace in September 2025.

The buying trap is identical on both sides: a partner agent has a license fee, and then it consumes your credits at the standard rate on top. A listing that looks cheap at $500 a month can cost several times that in metered consumption once real volume hits it. Before you buy any listing, ask the partner how many billable events a typical job produces. That number is your running cost and it is almost never on the listing page.

There is one Microsoft-specific restriction worth knowing if you are a builder rather than a buyer: declarative agents built in Copilot Studio and Agent Builder cannot be published to the Commercial Marketplace at all. We compare how each vendor's catalog is gated, and which license tiers can install from them, in our rundown of the major AI agent marketplaces.

How to actually choose between them

The platform decision is usually made before the pricing decision, and honestly it should be. If your revenue process lives in Salesforce, Agentforce agents sit on the record with the permission model already applied. If your work lives in Teams, Outlook and SharePoint, Copilot Studio agents sit where your employees already are and cost nothing extra for licensed users. Very few companies get to choose freely, and paying a premium to put agents where the data is not is rarely a good trade.

Where pricing should drive the decision is at the edges. Buy Copilot Studio if most of your agent work is employee-facing and you already hold Microsoft 365 Copilot licenses, because that usage is zero-rated and the arithmetic is not close. Buy Agentforce if your agents perform many small discrete operations per job, since one flat action charge beats five itemized ones, or if you need predictable per-conversation pricing that Microsoft does not sell. Watch the expiry clock if your volume is seasonal, because a monthly reset punishes spiky demand in a way an annual pool does not.

And weigh one option that neither vendor will raise. Both of these are platform commitments: a consumption contract, a seat mix, a connector project and an internal approval queue, entered before anyone knows how the agents will actually behave. If what you need is one agent doing one job well, a ready-made agent deployed on the stack you already run gets you to a working result without a meter you cannot mix, credits you cannot roll over, and a SKU swap standing between you and changing your mind.