Microsoft Copilot Cowork Goes GA: Model Choice Inside Office Is the Headline Nobody Is Writing
On June 16, 2026, Microsoft made Copilot Cowork generally available to every Microsoft 365 Copilot customer worldwide. Most coverage is focusing on the feature set. The story worth paying attention to is different: for the first time, users inside Microsoft Office can pick which AI model powers their work — Claude Opus, GPT-5.5, Sonnet 4.6, or Microsoft's own models. This is a structural shift in how enterprise AI is delivered.
What Copilot Cowork Actually Is
Copilot Cowork is not the chat interface that most people associate with Microsoft Copilot. It is Copilot's action mode — the version that does not just answer questions but takes multi-step actions inside Microsoft 365: drafting and sending emails on approval, creating meeting summaries and action items, updating SharePoint records, populating project plans in Planner, and initiating workflows.
Before GA, Cowork required a separate configuration and was available only through enterprise previews. Post-GA, it is on for every M365 Copilot customer by default. Over 50 percent of Fortune 500 companies were already running it in preview — which explains why Microsoft was comfortable with a broad GA rollout.
The Four Model Options and When to Use Each
- Claude Opus: Microsoft's current default for complex reasoning tasks — long document analysis, nuanced email drafting, multi-step planning. Anthropic's most capable model at the time of GA.
- Claude Sonnet 4.6: Faster, lower cost. Suitable for high-frequency tasks where quality requirements are high but not at Opus level — meeting summaries, routine communication drafts.
- GPT-5.5: Available in Frontier preview. Strong for multimodal tasks involving images and documents with mixed content types. The natural choice for organisations already deeply invested in OpenAI's ecosystem.
- Microsoft Cowork 1 (coming soon): Microsoft's own model, trained on M365 telemetry and optimised specifically for productivity suite workflows. Expected to lead on SharePoint and Teams integration depth.
Why Microsoft Made Claude the Default
This is the most interesting strategic question. Microsoft has a multi-billion dollar investment in OpenAI. Making Claude the default for complex tasks — ahead of GPT-5.5 — is a signal that deserves analysis.
The most likely explanation: Microsoft's internal evaluations showed Claude Opus outperforming GPT-5.5 on the specific task types that Cowork handles most frequently. Long document reasoning, instruction following for multi-step tasks, and nuanced drafting all appear to favour Claude in Microsoft's testing. The default is a quality signal, not a loyalty statement. It also reflects Microsoft's stated goal of being model-agnostic — they want customers to trust that Copilot recommends the best model, not the cheapest or the most strategically convenient one.
The New Usage-Based Pricing Layer
Cowork GA introduces a usage-based billing tier on top of the existing M365 Copilot subscription. The base subscription covers standard usage. High-volume agentic tasks — particularly those using frontier models like Claude Opus — consume Copilot Credits, which are billed separately.
This is the same playbook Anthropic used when they split their API pricing between standard and extended thinking models. Microsoft is establishing that frontier model capability has a consumption cost, not a flat-rate cost. For finance and procurement teams, this means M365 Copilot costs will be variable at scale, not fixed — a significant change from the previous per-seat billing model.
3 Things to Do This Week If Your Company Uses M365 Copilot
- Audit which Cowork tasks are running on which models: Log in to the M365 admin centre and review Copilot usage reports. Understand which task types are consuming which models and at what volume — this is your baseline before usage-based charges start accumulating.
- Define your model selection policy: Left to default, every user will use Claude Opus for everything. Define which task types should route to which models to control credit consumption. Routine summaries do not need Opus — Sonnet 4.6 is the right call.
- Test Cowork on your highest-value M365 workflow: Identify the one workflow that, if automated, would have the clearest ROI for your team. Run a structured test with Cowork this week. Measure time saved versus current process.
Key Takeaways
- Copilot Cowork is GA as of June 16 for all M365 Copilot customers — this is Copilot's action mode, not chat mode
- Users can now choose Claude Opus, Sonnet 4.6, or GPT-5.5 inside Microsoft Office — this is the first enterprise productivity suite to offer genuine model choice
- Microsoft defaulted to Claude Opus for complex reasoning tasks, not GPT — a quality signal, not an exclusivity arrangement
- Usage-based pricing on top of the M365 subscription means Copilot costs are now variable at scale — finance teams need to understand this before rollout
- Microsoft Cowork 1 (their own model) is coming and will likely lead on M365-native integration depth when it arrives
