Microsoft Copilot Studio (Copilot Credits Pricing, $200 per 25,000-Credit Pack)

Microsoft Copilot Studio is the company's low-code home for building AI agents and workflows, grown out of the old Power Virtual Agents product. What we're assessing here is less the canvas and more the commercial model attached to it, because that's where the supplied detail is richest — and where buyers are most likely to get caught out.

The headline is a $200-per-month capacity pack of 25,000 Copilot Credits, billed annually, with a pay-as-you-go alternative at a cent per credit through Azure. Authoring licenses are listed at $0, so the barrier to a first prototype is low, and agents can be pushed out to Teams, web chat, WhatsApp and custom apps once you're happy.

This will suit Microsoft 365 shops that already have Copilot seats and want internal assistants, since staff holding a $30-per-user-per-month license don't draw credits at all. It's a much more cautious proposition for anyone pointing an agent at external customers, where every interaction meters.

The pricing is coherent once you understand it, but it takes work to understand, and the economics swing wildly with design choices. A scripted FAQ answer costs a single credit; a generative answer costs two; a single reasoning-model response costs a hundred. That's a 100x spread hiding behind one toggle in a builder, and it means a well-meaning engineer can multiply a bill without ever touching the billing page.

Microsoft's own worked example makes the point better than we could: a support agent fielding 900 sessions a day burns 7,200 credits daily, which lands somewhere around $1,800 a month in packs. That is the real price of a customer-facing deployment, not $200.

The prepaid pack's effective $0.008 per credit is a genuine discount against pay-as-you-go — but only if you consume the lot, because unused pack credits don't roll over. That turns capacity planning into a guessing game at the start of an agent's life, and we'd be inclined to start on pay-as-you-go despite the premium until usage settles. The one-year P3 pre-purchase pool and its 5–20% tiers are a better deal still, though committing a year upfront to a workload whose credit burn you can't yet predict is a bet, not a saving.

Two further snags worth flagging: agents built on the GitHub Copilot harness start consuming credits during build, preview and test rather than after publishing, so experimentation isn't free on that path; and the credits sit on top of a stack of adjacent costs — Microsoft 365 licensing, Dataverse, Managed Environments, Azure compute and Azure OpenAI tokens for custom models. The sticker price is a floor, not a total.

We should be honest about the limits of this assessment. We have strong detail on commercials, channels and lineage, but little on build quality, model performance or how the authoring experience actually feels day to day — and the only user sentiment available is a 3.9-out-of-5 PeerSpot score drawn from just five reviews, which is far too thin to lean on. The trial also lets you build and test but not publish, so you can't fully validate before committing. Our rating reflects a flexible, well-tiered but genuinely opaque pricing model attached to a platform whose quality we can't yet independently vouch for.