M365 Copilot: Rollout of GPT-6.1 Sol and Claude Sonnet 5.5

On September 30, 2026, Microsoft kicked off the rollout of two new AI models in Microsoft Copilot with MC1483844: OpenAI’s GPT-6.1 Sol and Anthropic’s Claude Sonnet 5.5. This marks the second model expansion within a month, following the addition of Claude Opus 5.5 and GPT-6 Sol on September 22. Users will now select models via the model selector in supported Copilot interfaces.

The rollout is proceeding on two tracks. In Copilot Cowork and Copilot Studio, both models launched on September 30th with usage-based billing. In Word, Excel, PowerPoint, and Chat, they will be introduced in phases over the following days, within the scope of the Copilot user subscription license with usage limits. If a user reaches a limit, they will be notified and can switch to auto-mode or select a different model.

For administrators, the real news isn't the model selection itself, but the condition behind it. Claude Sonnet 5.5 only appears if the Anthropic subprocessor is enabled in the tenant. GPT-6.1 Sol, accordingly, depends on the release of the OpenAI subprocessor. If the respective switch isn't set or is deliberately disabled, the model simply won't show up. GPT-6.1 Sol runs on Microsoft's own Azure infrastructure, while the Claude models are integrated via the additional Anthropic subprocessor.

In the DACH region, this distinction is more than just a formality. Anyone who has deployed Copilot with GDPR compliance in mind has documented in their Record of Processing Activities which subprocessors are involved. A newly added subprocessor is therefore more than just a checkbox in administration—it is a matter for the Data Protection Impact Assessment, which must be addressed before the switch is flipped. Approval of the models thus belongs on the desk of the Data Protection Officer, not solely in IT.

Make a deliberate decision about whether and which of the new models to enable, rather than leaving the switches as they are. For each model, check the associated subprocessors status and reconcile it with your existing Copilot privacy documentation.

Address the cost aspect in parallel, as Cowork and Copilot Studio operate on a usage-based billing model, whereas Office apps have usage limits. Both factors should be communicated to the business units before the first users start exploring the new models.

Two new models in the selector may sound like a mere convenience update, but for administrators, they represent a governance decision. The pace at which Microsoft is rolling out models—two additions in a single month—makes it clear that approving AI models is not a one-time task but an ongoing process.

Once a clear process for subprocessor review, data protection alignment, and cost transparency is established, the next model addition becomes routine rather than a surprise.