Microsoft 365 Copilot adds GPT-6 Astra and Claude Fable 5.1 to the model selector: but they'll cost you credits

Two new AI models are coming to Copilot, and they sit outside your existing licence

From early November 2026, Microsoft 365 Copilot’s model selector is gaining two frontier models: GPT-6 Astra (from OpenAI) and Claude Fable 5.1 (from Anthropic). They’ll appear in Copilot Chat and in Copilot inside Word, Excel, and PowerPoint. The rollout is expected to complete by mid-November 2026.

The important distinction is how they’re billed. These are not included in the standard Microsoft 365 Copilot per-user licence. They run on usage-based billing, consumed through Copilot Credits. That changes what admins need to do before anyone can access them, and it changes how organisations need to think about AI spend.

This is covered under Microsoft Message Center item MC1519122, updated 4 November 2026.

What’s actually changing

Microsoft has been building towards a two-track billing model for Copilot for some time. The per-user licence covers the everyday AI experience: Copilot Chat, Copilot in Office apps, and the Auto model routing that picks the right model for each request based on accuracy, speed, and cost. That stays as it is.

Usage-based billing sits on top of that licence and covers more capable, more expensive models and workloads. GPT-6 Astra and Claude Fable 5.1 are the first frontier models to arrive in the model selector under this arrangement. When a user picks one of these models, the work is billed in Copilot Credits rather than covered by their seat.

One additional feature arrives with this rollout: when using a usage-based billing model in Copilot Chat, users will be able to set the reasoning effort level. That means they can dial up the depth of reasoning for a complex task, or keep it lighter for something straightforward. Naturally, higher reasoning effort will consume more credits.

Nothing happens automatically

This is worth emphasising clearly. Users will not suddenly see GPT-6 Astra or Claude Fable 5.1 in their model selector unless three things are true:

1. They have a Microsoft 365 Copilot user subscription licence.
2. Usage-based billing has been enabled for them by an admin.
3. Copilot Credits have been allocated to them.

For enterprise tenants, usage-based services stay off until an admin creates a spending policy in the Microsoft 365 admin centre. No spending policy means no billing, and no billing means no access to these models. Nothing is charged before an admin takes deliberate action.

Microsoft’s default spending limit, once UBB is enabled, is 4,000 Copilot Credits per licensed user per month. Admins can adjust that figure. Spending policies also allow admins to assign model access and reasoning effort levels by group, so it’s possible to restrict frontier model access to specific teams rather than opening it up organisation-wide.

If a user exhausts their allocated credits, they can request more from their tenant admin rather than being cut off silently.

What this means for IT admins

If you’re responsible for your organisation’s Microsoft 365 environment, this is primarily an admin story. The user experience change is real, but it won’t reach anyone until you act.

Before mid-November 2026, it’s worth working through the following:

Check licencing. Only users with a Microsoft 365 Copilot user subscription licence are eligible for usage-based billing models. Users without that licence won’t see the new options regardless of anything else.

Decide whether to enable UBB at all. There is no obligation to turn this on. If your organisation has no current need for frontier models, you can simply leave usage-based billing disabled. The standard Copilot experience continues unchanged.

If you do enable it, set spending policies deliberately. Think about which teams genuinely need access to GPT-6 Astra or Claude Fable 5.1 and size credit allocations accordingly. The ability to scope access by group is useful here, particularly if you want to run a controlled rollout rather than enabling it for everyone at once.

Communicate with users. If users will gain access to new models, a short explanation of what they are, how credits work, and how to request more if needed will save a wave of support tickets.

It’s also worth noting that the announcement does not specify exact credit consumption rates for GPT-6 Astra or Claude Fable 5.1. Those details will matter for forecasting spend accurately, so keep an eye on Microsoft’s pricing documentation as the rollout completes.

The bigger picture

This rollout is part of a deliberate structural shift in how Microsoft monetises Copilot. The per-user seat covers reliable, everyday AI. Frontier models with deeper reasoning and higher compute costs sit in a separate lane, billed by consumption. That approach lets Microsoft offer access to genuinely more capable models without bundling the cost into every seat regardless of usage.

For organisations, it means AI spend becomes more variable and more visible. The admin controls around spending policies and credit allocation are the mechanism for keeping that spend within bounds. Used well, they give meaningful governance over which teams access which capabilities and at what cost. Used carelessly, or not configured at all, the exposure is simply that frontier models remain inaccessible until someone takes action.

The underlying framework for all of this was set out in MC1479276, which established the Copilot Credits billing model. November 2026 is when that framework starts to matter practically, with real models appearing in real users’ model selectors for the first time.