Microsoft Copilot: All Models Available: productivity, platform tiers, and integration scope

Microsoft Copilot is not a single AI model but a unified framework that integrates several OpenAI-based and Microsoft-built large language models across productivity, enterprise, and developer environments. Depending on where it is deployed—Microsoft 365, Windows, GitHub, or Azure OpenAI Service—Copilot draws from different model families such as GPT-4 Turbo, GPT-4o, and OpenAI o-series variants fine-tuned for specific Microsoft ecosystems. Understanding the available models helps clarify differences in reasoning capability, context length, and connectivity between personal, professional, and enterprise tiers.

·····

.....

How Microsoft structures the Copilot model ecosystem.

Copilot’s architecture distributes models through layered access points rather than a single endpoint. Each surface—Word, Excel, Teams, Edge, Bing, and Azure—connects to the Copilot system through Microsoft Graph orchestration, which selects an appropriate model for each task.

There are four primary model families currently deployed:

1. GPT-4 Turbo and GPT-4o (OpenAI models hosted by Microsoft) – Used in Copilot for Microsoft 365, Bing Chat Enterprise, and the Windows Copilot runtime.

2. o3 and o4-series (OpenAI o-models) – Experimental reasoning and efficiency models introduced in Azure OpenAI and Copilot Pro environments.

3. Phi-3 family (Microsoft Research models) – Lightweight, highly efficient models used in Windows Copilot Runtime and Edge features for local reasoning and quick tasks.

4. Codex and GPT-4 code derivatives – Power GitHub Copilot and developer assistants within Visual Studio and VS Code.

Each model is orchestrated through the Copilot System Prompt layer, which governs user authentication, Graph data access, and compliance filters before the model executes the user query.

·····

.....

Models active in Microsoft 365 Copilot.

When Copilot is used inside Microsoft 365 apps—Word, Excel, PowerPoint, Outlook, and Teams—the back-end model is GPT-4 Turbo or GPT-4o, depending on tenant configuration.

App Surface	Model Used	Context Window	Notes
Word / PowerPoint / Outlook	GPT-4 Turbo	~128k tokens	Optimized for drafting and summarisation; handles long documents.
Excel Copilot	GPT-4 Turbo (with formula and data connectors)	~64k tokens	Integrated with Excel calculation engine; no external code execution.
Teams Copilot	GPT-4o	~100k tokens	Adds speaker recognition, recap, and transcript summarisation.
Microsoft Loop Copilot	GPT-4o	~100k tokens	Used for collaborative workspace summarisation and task generation.

In all cases, data grounding comes from Microsoft Graph, meaning the model only accesses data available to the signed-in user—emails, meetings, chats, or SharePoint files—without training on user content.

·····

.....

Models used in Bing Chat Enterprise and Windows Copilot.

The Bing Chat Enterprise and Windows Copilot experiences share the same core model family as Microsoft 365 Copilot but differ in orchestration and data access:

• Bing Chat Enterprise uses GPT-4 Turbo or GPT-4o, with web grounding through Bing Search API and citation tracing.

• Windows Copilot Runtime includes a compact Phi-3 Mini / Phi-3 Small model for quick, local reasoning and summarisation tasks.

• In newer Windows builds, the Copilot Runtime API allows system apps to call a local Phi-model first, escalating to cloud GPT-4o for complex reasoning.

Phi-3 models, trained by Microsoft Research, offer small-scale on-device inference with strong efficiency-to-accuracy ratios, supporting basic code explanations, settings recommendations, and summarisation of local content without network latency.

·····

.....

Models available in Azure OpenAI Service and Copilot Studio.

Enterprises that build their own Copilot extensions or deploy AI internally interact with Copilot through Azure OpenAI Service and Copilot Studio. These environments expose the full set of current OpenAI models hosted within Azure’s compliance boundary.

Azure / Studio Model Name	Purpose	Typical Context Limit
gpt-4-turbo	Default Copilot model for reasoning and generation	128k tokens
gpt-4o	Multimodal (text, image, audio) reasoning	128k–1M tokens depending on deployment
o3-mini / o3-pro	Optimised reasoning-efficiency models	32k–200k tokens
gpt-35-turbo	Legacy compatibility tier	16k tokens
phi-3-mini / phi-3-medium	Lightweight Microsoft Research models for local or serverless Copilots	Variable, typically ≤8k tokens

Copilot Studio combines these with Power Automate, Graph Connectors, and Plugins. Developers can build custom Copilot actions using function-calling patterns identical to OpenAI’s tool-calling schema.

·····

.....

GitHub Copilot and developer model family.

GitHub Copilot, integrated in Visual Studio, VS Code, and JetBrains IDEs, primarily relies on GPT-4 Turbo (code-tuned). A small subset of features still use Codex derivatives for autocompletion latency reasons, but Microsoft continues migrating toward GPT-4o Coder in 2025.

Developer Context	Model	Focus
Autocomplete / inline suggestions	Codex Hybrid → GPT-4o Coder	Fast local completions
Chat and explanations	GPT-4 Turbo	Contextual reasoning
Test generation / refactoring	GPT-4o Coder	Multimodal (code + diagrams) support

These models interact with repositories under strict privacy gating—Copilot reads only files open in the session and does not transmit source code for training.

·····

.....

Comparative framework — Copilot model distribution by platform.

Copilot Platform	Core Model(s)	Special Capability
Microsoft 365 Apps	GPT-4 Turbo / GPT-4o	Document reasoning, Graph grounding
Teams / Outlook	GPT-4o	Speaker recognition, meeting recap
Bing Enterprise	GPT-4 Turbo	Web grounding, citations
Windows Copilot	Phi-3 Mini + GPT-4o	Local + cloud orchestration
GitHub Copilot	GPT-4o Coder	Code reasoning and autocompletion
Azure OpenAI Copilot Studio	gpt-4o / o3 / o4-mini	Custom agent development

This layered ecosystem lets Microsoft dynamically route requests to the most efficient model, reducing cost and latency while preserving enterprise-grade compliance.

·····

.....

How users can identify which model their Copilot uses.

• In Bing Chat or Edge Copilot, typing “What version of GPT are you using?” displays whether it runs GPT-4 Turbo or GPT-4o.

• In Copilot Studio, the “Model Picker” explicitly lists available engines (e.g., gpt-4-turbo, gpt-4o, o3-mini).

• In Microsoft 365 Copilot, model choice is managed by Microsoft Graph Control Plane and not user-selectable; documentation specifies that GPT-4 Turbo or newer applies.

• In Azure OpenAI, the endpoint name (model = "gpt-4o" etc.) determines which engine handles the request.

Users in enterprise tenants often receive transparent upgrades when Microsoft switches to newer OpenAI models globally.

·····

.....

Security, privacy, and compliance layers across models.

All Copilot instances—regardless of model—run inside Microsoft’s commercial compliance boundary with isolation from consumer endpoints. Content processed by Copilot for Microsoft 365, Bing Enterprise, and GitHub Copilot Business is not used for model training.

Each request passes through:

1. Graph grounding service, which restricts data to what the user can access.

2. Compliance and logging layer, enabling audit trails and eDiscovery.

3. Safe Prompt Orchestration, preventing data leakage across models.

4. Regional data storage in applicable enterprise tenants (EU Data Boundary support).

This layered control allows Microsoft to integrate frontier models while maintaining regulatory compliance across jurisdictions.

·····

.....

Feature evolution and upcoming model transitions.

• 2024–2025 updates: Gradual shift from GPT-4 Turbo to GPT-4o as the default model in Microsoft 365 Copilot and Bing Enterprise.

• Azure OpenAI updates: Introduction of o4-mini for low-latency Copilot Studio actions.

• Windows Copilot Runtime: Continuous optimisation of Phi-3 Small for on-device tasks and hybrid cloud routing.

• GitHub Copilot X roadmap: Expansion toward GPT-4o Coder for multimodal developer workflows (code + diagrams + documentation).

Future Copilot generations will dynamically mix local Phi-models and cloud GPT-4o reasoning to balance performance, privacy, and cost.

·····

.....

Operational recommendations for enterprises and developers.

• Identify the tenant configuration (Business Standard, Enterprise E5, or Education) to confirm which Copilot tier and model apply.

• Use Copilot Studio to test custom actions under different models (e.g., compare gpt-4-turbo vs o3-mini).

• For hybrid workloads, deploy Phi-3 on Windows devices to handle local summarisation while routing complex tasks to cloud GPT-4o.

• Monitor the Microsoft 365 Admin Center for release notes announcing backend model transitions.

• Train internal users to phrase prompts explicitly, as each model variant optimises for different verbosity and latency.

Understanding which model is in play—GPT-4o, Turbo, o-series, or Phi-3—allows enterprises to tune Copilot’s behaviour for both compliance and performance across Microsoft’s expanding AI stack.

·····

.....

FOLLOW US FOR MORE

DATA STUDIOS

.....[datastudios.org]