GPT-5.6 is now the preferred model for Microsoft 365 Copilot across Word, Excel, PowerPoint, and Cowork

OpenAI has named GPT-5.6 the preferred model powering Microsoft 365 Copilot, bringing the latest model family into Word, Excel, PowerPoint, Copilot Chat, and the new collaborative feature called Cowork. For AI builders shipping enterprise integrations or productivity agents, this designation signals how large model publishers manage model exposure, licensing, and deployment paths for enterprise customers. The announcement also lands amid ongoing questions about the Microsoft-OpenAI relationship as Microsoft builds its own MAI models for cost control.

What happened

GPT-5.6 is a family of three models: Sol, Terra, and Luna, designed for flagship, enterprise, and high-volume knowledge-work use cases. Microsoft will access the models both natively within Copilot and through the OpenAI API.

In Word, Copilot is designed to turn rough ideas into more complete drafts with fewer prompting rounds. In Excel, it supports more complex analysis with less manual assembly. In PowerPoint, Copilot generates richer presentation drafts with better visual balance. In Cowork, the model carries multi-step tasks from initial instruction to finished result, rather than returning a draft for the user to complete.

The model received broad U.S. regulatory clearance from the Commerce Department's Center for AI Standards and Innovation, ending a period of restricted access.

Why AI builders should care

For enterprise AI builders and product teams, GPT-5.6 for Microsoft 365 Copilot demonstrates a key pattern: how model publishers designate a preferred model for a major productivity suite while leaving room for multi-provider strategies. Microsoft has been building its MAI model family for coding, image generation, transcription, and voice, and has also used Anthropic models in certain Copilot tasks after internal testing. The company hosts models from Meta, Mistral, and other providers in its data centers.

This matters for builders because the "preferred model" designation doesn't mean exclusive use. As TechCrunch noted, OpenAI's software can hold that label while Microsoft continues building its own AI capabilities to manage costs. If you're building on Microsoft's AI stack, you should expect continued model variety and evolving default selections.

Practical implications

Organizations can access GPT-5.6 in Copilot via native integration or the OpenAI API, depending on tenant configuration and region. Users may see GPT-5.6 selected automatically by Copilot when the system determines it fits the task, or they can choose it from a model selector where offered.

For teams building knowledge-work agents in regulated environments, the Commerce Department clearance is a signal that the model has passed government evaluation for broader deployment. That could simplify compliance discussions for enterprises needing approved models.

For developers integrating via API, the GPT-5.6 family offers tiered capabilities through Sol, Terra, and Luna, letting you match model sophistication to task cost. The native Copilot path, meanwhile, abstracts model selection behind Microsoft's routing logic, which may simplify deployment but reduce cost control.

Caveats

The source context includes important limitations. Availability may vary by region and tenant configuration. No performance metrics, pricing, or token limits for the GPT-5.6 family are provided in the cited sources. The model variants Sol, Terra, and Luna are described by use-case tier but without benchmark comparisons or capability boundaries. The long-term balance between OpenAI and MAI models remains unresolved, and Microsoft's cost-control incentives could shift preferred model designations over time.