More Models, One Copilot: Claude Opus 5.5 and GPT-6 Sol in Microsoft Copilot
On September 22, 2026, Microsoft began rolling out two new frontier models in Copilot: Anthropic's Claude Opus 5.5 and OpenAI's GPT-6 Sol. They are arriving in Word, Excel, PowerPoint, Copilot Chat, Cowork, and Copilot Studio. Microsoft's reasoning: no single model is ideal for every task. But response quality depends not only on the model — it also depends on what Copilot knows about a specific company's work.
What was announced
Both models, Claude Opus 5.5 and GPT-6 Sol, are arriving in Copilot at the same time — not just in a standalone chat, but across all the main work scenarios:
- Word, Excel, PowerPoint — working with documents, spreadsheets, and presentations;
- Copilot Chat — the conversational assistant for everyday requests;
- Cowork — agent scenarios where Copilot carries out multi-step work;
- Copilot Studio — building the company's own AI agents.
The rollout started on the day of the announcement. Microsoft notes that availability may vary by license, access, and region. Check the AI at Work Roadmap and the Microsoft Copilot release notes for the current status.
Why Microsoft is betting on multiple models
A few years ago, Copilot was mostly associated with OpenAI models. Now Microsoft is steadily expanding the choice: different models have different strengths, and the company wants customers to get new AI capabilities as soon as they emerge, without leaving the corporate environment. For businesses, this means there's no need to buy separate subscriptions to different AI services or spread company data across external platforms just to use the strongest model for a specific task.
We've already covered how Copilot differs from external AI chats in Microsoft Copilot vs ChatGPT: what's the difference for business. With this announcement, the difference becomes even clearer: the question is no longer which model to choose, but where that model works with company data.
The model is only half the answer: the role of Work IQ
In the same announcement, Microsoft highlights a second quality factor. Through Work IQ, Copilot grounds its responses in the organization's files, meetings, chats, and business data. That way, the answer relates to the user's actual work, not just the model's general knowledge.
In practice, this matters more than it seems. The most powerful model without access to context gives a generic answer; a model with access to client correspondence, meeting notes, and a CRM report gives a specific one. So even with Claude Opus 5.5 and GPT-6 Sol available, Copilot's results in a company will largely depend on how well-organized the data in Microsoft 365 and business systems is.
What changes for teams
Documents and analytical memos
For long texts, contracts, and analysis, you can pick the model that handles that type of task best — without leaving Word.
Spreadsheets and calculations
In Excel, the new models work right next to the data in the workbook: analysis, formulas, explaining variances in financial and sales reports.
Custom agents in Copilot Studio
Agent builders get a wider choice of models for business scenarios: handling requests, preparing proposals, internal help-desk bots.
Agent scenarios in Cowork
Multi-step tasks that Copilot performs on its own can now run on the new frontier models.
What IT and leadership should check
1. Availability in your tenant. Microsoft states that access depends on license, access level, and region. Don't promise employees the new models until you've confirmed they're available in your environment and in the Microsoft 365 admin center.
2. Policy on third-party models. Claude Opus 5.5 is an Anthropic model — a third-party provider from Microsoft's perspective. Your security team should review the data processing terms for such models separately and decide which users and scenarios to enable them for. We described the basic principles of data protection in Copilot in Copilot security and confidential company data.
3. Costs of agent scenarios. Cowork and Copilot Studio agents are among the scenarios billed through Copilot Credits. Starting November 2, 2026, new Copilot Business licenses in CSP automatically get usage-based billing (more in Copilot Credits in CSP from November 2026). The more powerful the models available to agents, the more important it is to set up budgets and consumption monitoring in advance.
4. User training. Model choice only makes sense when people understand what it's for. Without short internal guidelines like "which model to use for which task", most users will stick with the default.
Our recommendation
For companies that already have Microsoft 365 Copilot, the arrival of Claude Opus 5.5 and GPT-6 Sol is a reason to test the new models on 2–3 real work scenarios — for example, drafting an analytical memo, breaking down a financial spreadsheet, and an agent in Copilot Studio — and compare the results with the default model. That way you get your own data on which model works best for your tasks, instead of relying on generic benchmarks.
For companies still choosing an AI tool, the announcement shows where the market is heading: instead of betting on a single model, a platform where models change and improve, while data, access rights, and integrations with business systems stay in one managed environment.
FAQ
Where exactly in Microsoft Copilot are Claude Opus 5.5 and GPT-6 Sol available?
According to Microsoft's announcement of September 22, 2026 — in Word, Excel, PowerPoint, Copilot Chat, Cowork, and Copilot Studio. The rollout is gradual, and availability depends on license, access, and region.
Do I need a separate Anthropic or OpenAI subscription to use these models in Copilot?
No. The models become available inside Microsoft Copilot, so separate Anthropic or OpenAI accounts aren't needed. That said, agent scenarios such as Cowork and Copilot Studio agents may be billed through Copilot Credits.
Which model should I choose?
Microsoft doesn't give a universal recommendation and explicitly says no single model is ideal for every task. The most reliable approach is to compare models on a few typical tasks your team handles and capture the results in internal guidelines for users.
Does the choice of model affect which company data Copilot can see?
Copilot gets context for its responses through Work IQ — from the organization's files, meetings, chats, and business data, respecting the user's access rights. Before enabling third-party models, IT should separately review the data processing terms that apply to them.
Want more from Copilot than just new models?
Progresia helps configure Microsoft 365 Copilot around your company's processes: data readiness audits, access policies, Copilot Credits cost control, and agent development in Copilot Studio.