Model Choice in Copilot Cowork: From AI Access to a Model Portfolio
Copilot Cowork now offers several models, depending on what the organization allows: Auto, Claude Sonnet 5, Claude Opus 4.8, GPT 5.5 (Frontier) and a paired Sonnet + Opus Advisor mode. A quick meeting summary needs different capabilities than a high-stakes research synthesis. Microsoft makes this choice available inside enterprise controls, with visible retention notices and admin settings per model family. In this article, I explain why enterprises move from AI access to managing a model portfolio and what IT leaders should watch.
Microsoft is introducing explicit model selection in Microsoft 365 Copilot experiences such as Copilot Cowork. Users and administrators choose different models depending on the task and remain inside the governed Microsoft environment. With this step, enterprises move from AI access to managing a portfolio of models.
Which models are available
According to Microsoft Learn, Copilot Cowork offers these options, depending on organizational settings:
- Auto as the default for most daily work
- Claude Sonnet 5 for everyday tasks and fast responses
- Claude Opus 4.8 for complex, high-stakes work with deeper reasoning
- GPT 5.5 (Frontier) as a versatile model, strong for longer texts and citations
- Sonnet + Opus Advisor as a paired mode in which one model handles the main turn and the other reviews it for accuracy and completeness
Different work needs different models
A meeting summary, an email draft or a short internal note benefits from speed. A strategic memo or a research synthesis justifies a slower, more careful model. Copilot's value will therefore depend increasingly on how well the platform routes each task to a suitable reasoning profile.
This changes the questions IT leaders ask: Which model fits drafting, and which fits complex analysis? Which models meet our compliance requirements? Where do prompts and responses remain? How much should users decide, and how much administrators?
Governance made visible
Microsoft does not treat model choice as a pure performance question. Some preview models require data retention by the model provider. Copilot shows this with notices in the model picker and banners in the product. Administrators control the availability of model families in the Microsoft 365 admin center.
Adoption often slows at the gap between what a model can do and what a company will allow. Visible retention notices and admin controls turn model choice into a decision that the organization can govern.
Consequences for Microsoft's position
Orchestration over model ownership Enterprises do not need loyalty to one model family. They need reliable results under governance. Microsoft can compete through orchestration and integration instead of claiming that one model fits every scenario.
A more valuable control plane With more models, the administrative layer gains weight. Organizations have to decide who may use which model, what is enabled by default and which models are suitable for regulated processes.
Transparency for users When users see which model produced an answer and whether data handling changes, they understand the system better.
Deliberate workflow design Organizations can plan fast models for routine drafts, deeper reasoning for executive preparation, paired models for important deliverables and restricted availability for sensitive functions.
Simple for users, detailed for governance
Auto as the default shows the principle. Most users should not have to think about model selection. The platform should still let organizations introduce specialization where it improves results. If every user has to benchmark models, adoption stalls. If organizations cannot control availability, retention and task fit, trust stalls.
As this area develops, I would watch five questions. Is the model choice intuitive for users? Does Microsoft communicate differences in data handling clearly? How detailed are the admin controls? Can organizations map model selection to workflow types? Does specialization measurably improve work quality? The more of these questions Microsoft answers convincingly, the more model choice becomes a strategic layer of its AI offering.