Copilot Studio is priced by consumption: a capacity pack of 25,000 Copilot Credits (messages) costs $200 per tenant per month on an annual commitment — roughly $0.008 per credit — or pay-as-you-go at about $0.01 per credit with no upfront commitment via an Azure meter. Each interaction burns a variable number of credits: a classic answer is ~1 credit, a generative answer ~2, tenant-graph grounding ~10, and an autonomous agent action 25 or more. The real cost is set by agent architecture, not the headline pack price.
Copilot Studio has no per-seat price. It is metered in Copilot Credits, and you acquire that capacity two ways: a capacity pack of 25,000 credits for $200 per tenant per month on an annual commitment (about $0.008 per credit), or a pay-as-you-go meter billed through an Azure subscription at roughly $0.01 per credit with no commitment. Microsoft 365 Copilot seats at $30 per user per month include a built-in allowance for using packaged agents, but custom agents — especially autonomous ones — draw down Copilot Credit capacity that is invoiced separately. The number that matters is not the pack price; it is how many credits each agent interaction consumes.
| Purchase model | Unit | List price | Effective per credit | Commitment |
|---|---|---|---|---|
| Capacity pack (prepaid) | 25,000 Copilot Credits | $200 / tenant / mo | ~$0.008 | Annual |
| Pay-as-you-go meter | Per credit consumed | ~$0.01 / credit | ~$0.010 | None (Azure) |
| Included with M365 Copilot | Packaged-agent allowance | Within $30 / user / mo | n/a | Copilot seat |
| Additional packs (scale) | 25,000 credits each | $200 / pack / mo | ~$0.008 | Annual |
A credit is the billing atom, and consumption is the part that surprises budget owners. A single user message can cost anywhere from 1 to 200+ credits depending on how the agent is built. The published 2026 rates we model against in client deals are below — note that one "conversation" frequently chains several of these events, so a 10-turn support session grounded in your tenant data can burn 100–150 credits, not 10.
| Interaction type | Approx. credits | What triggers it |
|---|---|---|
| Classic answer | ~1 | Topic/keyword response, no generative AI |
| Generative answer | ~2 | AI-generated response from a knowledge source |
| Tenant graph grounding | ~10 | Answer grounded in Microsoft 365 tenant data |
| Agent action / connector | ~5–25 | Calling a Power Platform or external action |
| Autonomous agent action | 25+ | Event-triggered, multi-step autonomous run |
Credit burn, not seat count, sets the bill. We model agent-level consumption before you commit to packs.
Pay-as-you-go is the right entry point for pilots and unpredictable workloads — no commitment, and you only pay for credits consumed at month end. Capacity packs are about 20% cheaper per credit ($0.008 versus $0.01) and make sense once monthly consumption is stable enough to forecast. In our engagements the disciplined pattern is to launch every new agent on the pay-as-you-go meter, collect 60–90 days of telemetry from the Copilot Studio admin analytics, then convert the steady-state baseline to annual packs while leaving spiky or seasonal agents on the meter. Buying packs first — before you know your burn — is how organisations end up paying for 25,000 credits a month they never touch, or topping up at the higher meter rate because one pack was not enough.
The structural lever most buyers miss: Copilot Studio capacity is increasingly negotiated inside the broader Microsoft agreement rather than bought à la carte. If your Enterprise Agreement renewal is on the horizon, Copilot Studio credits, Microsoft 365 Copilot seats, and the underlying base SKUs should be priced as one package — that is where the concessions live. For the full seat-side picture see our Microsoft Copilot pricing guide, and for how the agent licensing rules fit together, the Copilot Studio licensing breakdown.
Seat-selection framework, agent-economics benchmarks, and the commit structure that keeps credit spend honest.
Credit overruns are an architecture problem, not a pricing problem. The agents that blow budgets are the ones grounded in tenant data on every turn and chaining multiple autonomous actions where a single generative answer would do. Across our 340+ enterprise engagements the controls that hold spend down are consistent: cap grounding to the turns that need it, set per-agent capacity limits in the admin centre, route high-volume FAQ traffic to classic (1-credit) answers, and review the consumption analytics monthly so a misconfigured agent is caught in week one, not at the quarterly invoice. Treat each agent as a cost centre with a credit budget, and the $200 pack stretches far further than a tenant-wide free-for-all ever will.
The licensing structure also matters: Microsoft 365 Copilot seats carry a packaged-agent entitlement, so workloads that fit inside that allowance should stay there rather than spilling onto metered Copilot Studio capacity. Mapping which use cases belong on the seat versus on credits is the single biggest line-item lever, and it maps directly back to your Microsoft licensing baseline and how the Microsoft practice structures the wider estate. Where the agent strategy touches the contract, our contract negotiation team prices the credits and the seats together.
We size credit consumption and contract terms together — the two numbers that set the return before you sign.
For Copilot Studio commitments above $250K annually, independent advisory across agent economics and commit structure typically returns several times the fee. In our experience the credit-burn forecast is where the money is made — not the headline pack price. We have benchmarked Microsoft AI deals from single-agent pilots to tenant-wide rollouts, and the pattern holds: model the consumption first, then negotiate the capacity. For broader context on how AI tools are priced across vendors, see our AI vendor pricing benchmarks, and the proof points across recent engagements in our case studies.
We have run Microsoft AI negotiations from single-agent pilots to tenant-wide deployments.
Weekly compliance intelligence for IT leaders.
