Draft

Last updated on

Copilot vs Copilot Studio: what you're actually paying for

If you’ve been asked to budget for Microsoft Copilot this year, you’ve probably run into two very different pricing models hiding under one brand name. Here’s the distinction that actually matters.

Microsoft 365 Copilot is a per-user licence. You add it to an existing Microsoft 365 plan, and it lights up AI features across Word, Excel, Teams and the rest of the Office suite, plus Copilot Chat. Simple, predictable, seat-based.

Copilot Studio is a different animal entirely. It’s a low-code platform for building custom AI agents that plug into Teams, SharePoint, Power Platform, external CRMs and APIs. And critically, it isn’t licensed per seat. You don’t pay per user, per conversation, or per agent. You pay for consumption, measured in Copilot Credits.

Here’s how that consumption works: every time a user interacts with an agent, or an agent performs a task on their behalf, it draws down credits. Credits measure the effort involved in retrieving information, responding to prompts, and executing actions such as workflows or tool calls. A simple lookup costs less than a multi-step task that touches several systems — a classic answer runs about 1 credit, a generative answer 2, an agent action 5, and grounding against the tenant graph 10.

You can access these credits through a few different channels: a pay-as-you-go meter, a subscription-based credit pack, or one of Microsoft’s pre-purchase plans. Credit packs are sold in blocks of 25,000 credits for $200/month, and unused credits don’t carry over, so it’s worth sizing a subscription against real usage rather than rounding up “to be safe”. If you go the subscription route, it’s worth also setting up the pay-as-you-go meter as a safety net: with it linked, overage bills to your Azure subscription instead of getting you cut off.

The good news for anyone already paying for Microsoft 365 Copilot: since late 2025, that licence has included full access to Copilot Studio’s features for internal use, at no extra cost. But the condition for that zero-rating is narrower than “internal versus external” — it only covers agents that run under the authenticated M365 Copilot user’s own identity, like a colleague asking an agent a question directly. An internal agent that triggers itself automatically still burns credits even though it never leaves the building: Microsoft’s own worked example is an order-triggered internal agent costing 20 credits a day. The same carve-out applies to agent flows, where only runs started by the “When an agent calls the flow” trigger are covered — every other trigger is billed at the standard rate. Computer-Using Agents are excluded outright, and the whole inclusion sits behind fair usage limits Microsoft can revise.

The practical takeaway for anyone planning a Copilot Studio rollout:

  • User-invoked agents running under a licensed person’s identity: likely already included in your M365 Copilot spend
  • Autonomously triggered agents, Computer-Using Agents, and anything running at scale or for external users: budget for Copilot Credits separately
  • Set up usage monitoring early. Enforcement kicks in at 125% of prepaid capacity, at which point custom agents are disabled rather than throttled back — credit consumption is much harder to forecast retroactively than to track from day one

Microsoft’s own licensing guidance for this space has shifted several times over the past year, so if you’re scoping a serious deployment, it’s worth checking the latest guide rather than working from what you heard six months ago.

Have you hit surprise credit costs scaling a Copilot Studio agent? Curious how others are budgeting for this.