Copilot Model Map — Which AI Model Powers Each M365 Copilot Feature
Interactive map showing which AI models (GPT-5, Claude, Phi-4, o3-mini) power each Microsoft 365 Copilot feature. See multi-model orchestration patterns, admin controls, and routing logic.
Features: BizChat, Copilot in Word, Excel, PowerPoint, Outlook, Teams, Researcher Agent, Analyst Agent, Copilot Cowork, Copilot Studio, Agent Mode, Copilot Pages.
Models: OpenAI GPT-5.4, GPT-5.3, GPT-4o, Anthropic Claude Opus 4.6, Claude Sonnet 4, Microsoft Phi-4, OpenAI o3-mini.
Copilot Model Map
Which AI model powers each M365 Copilot feature — GPT, Claude, Phi
Which AI model powers each M365 Copilot feature — GPT, Claude, Phi
| Feature |
|---|
Click any cell to see details. Hover a column header to highlight all features using that model.
Click any feature or model to highlight its connections
Microsoft 365 Copilot doesn't rely on a single AI model. It uses multiple models together in sophisticated patterns. Here's how.
What can you actually control as an M365 admin? Here's the truth — most model routing is automatic, but there are key levers you can pull.
Copilot Studio — Model Selection
Agent creators can explicitly choose which AI model powers their custom agent. Options include OpenAI GPT-5.x, Claude Opus, Claude Sonnet, and GPT-4o.
Multi-Model Opt-In
Tenant admins can enable or disable Anthropic Claude models for Copilot Chat and Researcher. When enabled, Copilot can auto-route to Claude for suitable tasks.
Frontier Programme
Opt-in programme for advanced multi-model features like Researcher's Critique and Council modes. Requires admin approval.
In-App Copilot (Word, Excel, PowerPoint, etc.)
Microsoft manages per-request model routing automatically. However, admins control whether Anthropic models are enabled at tenant level — which affects Agent Mode and advanced features in Office apps.
Cowork & Analyst
These built-in agents use fixed model assignments (Claude for Cowork, o3-mini for Analyst). Admins can control availability — deploy Cowork to all users, specific groups, or block it entirely.
Building a custom agent in Copilot Studio? Answer 4 quick questions and we'll recommend the best model for your use case.
When did Microsoft change the AI models behind Copilot? A timeline of key model changes and milestones.
Which AI models does Microsoft 365 Copilot use?
M365 Copilot uses a multi-model strategy including OpenAI GPT-5.x, Anthropic Claude Opus and Sonnet, Microsoft Phi-4, and OpenAI o3-mini. Different features use different models — Copilot automatically routes to the best model for each task.
Can admins control which AI model Copilot uses?
It depends on the feature. Copilot Studio gives full model choice to agent creators. BizChat allows tenant-level opt-in for Anthropic models. Most in-app Copilot experiences (Word, Excel, Teams) have no admin model control — Microsoft manages routing automatically.
What is Critique mode in Copilot Researcher?
Critique mode is a multi-model pattern where GPT drafts the research response and Claude independently reviews it for accuracy, completeness, and citation quality — similar to academic peer review. This is the default mode for Researcher.
Does Copilot Cowork use Claude or GPT?
Copilot Cowork primarily uses Anthropic Claude as its reasoning engine for deep planning and multi-step execution. GPT may handle specific sub-tasks within Cowork workflows.
Is this data official from Microsoft?
We source all data from official Microsoft blog posts, Microsoft Learn documentation, and public announcements. Each relationship is tagged as 'official' or 'inferred' with a verification date so you know the confidence level.
Data last verified: 29 April 2026 · Auto-checked on the 1st & 15th of each month · Sources: Microsoft Blog, Microsoft Learn, Tech Community · See something outdated? Let us know