Microsoft splits Copilot pricing: a fixed licence for everyday work, a meter for agents

What changed

On 25 September 2026 Microsoft announced a rebuilt Copilot and, in a companion post, a two-part pricing model. The product changes are three new capabilities: Home, which brings Chat and Cowork into one starting point with Word, Excel and PowerPoint built in; Code, which lets non-developers describe an app, dashboard or automation and have Copilot build and host it inside the tenant; and Autopilot, a persistent agent with its own identity and workspace that keeps working between prompts.

The pricing change is the part to act on. Microsoft now divides Copilot into everyday AI and advanced AI. Everyday AI, meaning chat and Copilot in Word, Excel, PowerPoint, Outlook and Teams, stays on the per-user subscription licence, with model selection included. Microsoft says GPT-5.6 and Claude Sonnet 5 are in the licence and Claude Opus 5 will be included with limits. Advanced AI, meaning Cowork, Code, Autopilot, new advanced experiences in apps such as SharePoint, and the frontier models Microsoft names as Fable and Astra, runs on usage-based billing with Copilot Credits. The licence is required to reach any of it.

Microsoft published no credit price in either post. The only description of the rate is that Cowork costs vary with the model, the context, the tools used and the runtime. What it did publish is the control surface. For enterprise customers, usage-based services stay off until an administrator creates a spending policy in the Microsoft 365 admin centre, and nothing is billed before that. A policy sets budgets at tenant and group level, per-user caps and alert thresholds, and decides what a user sees on reaching a limit. Administrators can restrict model families by group, assign usage to Azure subscriptions or resource groups, and manage policies through the Microsoft Graph API. The licence itself carries fair-use limits: a heavy user gets a warning, then a choice between switching to automatic model routing at no extra cost or moving to credits.

Availability is staged. Home and Code are rolling out through Microsoft's Frontier programme, Autopilot is in private preview, and Code reaches Microsoft 365 Premium and Pro subscribers in preview later this year.

Why it matters for UK business

Microsoft has adopted the shape the rest of the market already had: a flat seat for conversation, a meter for delegation. That matters because delegation is where the spend is moving. Our May briefing on why bills rise while token prices fall made the general point; this is the same mechanism arriving inside the Microsoft 365 estate that most UK firms already own. A Copilot budget now has a predictable line and a variable one, and the variable one scales with how much work people hand to agents rather than with headcount.

The default is the safe one for enterprise customers, and that puts the first decision with the administrator, not the user. Until a spending policy exists, nobody can run up a bill. Once one exists, its contents are the control: who has a budget, how large, on which models, and what happens at the limit. The posts state this default for enterprise customers. They do not say how it behaves on small-business plans, so a smaller tenant should check its own admin centre rather than assume.

For a UK tenant there is a second decision folded into the first. Some metered capabilities use Anthropic models, and Microsoft's documentation states that Anthropic models in Microsoft 365 Copilot are disabled by default for UK tenants at the subprocessor setting and sit outside the EU Data Boundary. Our article on Anthropic models in Microsoft 365 Copilot sets out those defaults. Enabling a workload that depends on them is a data-processing decision as well as a cost decision, and the terms for a frontier model may not match the terms for the default ones. Confirm them with Microsoft for the specific workload.

Two cautions on the numbers. Microsoft's post includes a cost comparison showing Copilot against Claude and ChatGPT. It is Microsoft's own internal analysis, modelled on list prices across 40 prompts, and it priced Claude on Opus 5. Anthropic released Opus 5.5 at $4 and $20 per million tokens on 22 September, three days earlier, so the comparison was dated on arrival. Second, Microsoft says some capabilities will start on the meter and move into the licence as the economics allow. Early adopters will pay per use for things that later become included. That is a reasonable price for going first, provided someone chose to pay it.

What to do, and what not to do

Do:

• Write the spending policy before anyone asks for Cowork, Code or Autopilot: a tenant budget, group budgets, per-user caps and alert thresholds, and a decision on what users see at the limit.
• Restrict model families by group. Automatic routing suits most people; reserve frontier models for teams that can state what they need them for.
• Run one delegated workflow with a small, capped group for a month and read the consumption reporting before widening access. With no published rate, measurement is the only forecast available.
• Map usage to cost centres through Azure subscriptions or resource groups from day one, so agent spend appears beside the rest of each team's budget.
• Review the Anthropic subprocessor setting and update the DPIA before enabling any workload that depends on Anthropic models.
• For your heaviest agentic users, compare total cost with the alternatives, including Claude Team and Enterprise, on your own workload rather than a vendor chart.

Do not:

• Budget from Microsoft's comparison. It is vendor analysis on superseded prices.
• Assume the per-user licence covers agents. It covers chat and the Office apps; delegated work is metered.
• Plan a rollout around features that are still in the Frontier programme or private preview.

Where The AI Consultancy fits

Deciding which work belongs on a flat licence, which justifies a meter and which vendor should carry it is a strategy question before it is a configuration one. We take no referral or reseller income from Microsoft, Anthropic or OpenAI, as set out on our independence page, and our AI strategy service works through that comparison on a client's own usage.

Verified on 10 October 2026 against Microsoft's "Introducing the new Copilot with Home, Code and Autopilot" (25 September 2026, modified 2 October 2026) and "Evolution of the Copilot pricing model" (25 September 2026, updated 28 September 2026), and Anthropic's "Introducing Claude Opus 5.5" (22 September 2026). Microsoft published no credit price in either post. This briefing is general information, not procurement or data-protection advice; confirm current terms and availability for your tenant with Microsoft.