Microsoft Copilot All Models Available: orchestration, providers, and on-device options

Microsoft Copilot does not rely on a single large language model but instead orchestrates multiple providers and model families depending on the surface where it is used. Across Microsoft 365, GitHub, Bing, Windows Copilot+ PCs, and Copilot Studio, the system combines OpenAI’s GPT models with Anthropic’s Claude, Google’s Gemini in some coding contexts, and even Microsoft’s own Phi small models for on-device acceleration. In 2025, the catalog of available models expanded significantly as Microsoft opened Copilot Studio to external model selection and brought new defaults into GitHub and Visual Studio.

·····

.....

How Microsoft 365 Copilot is powered.

In Microsoft 365 apps such as Word, Excel, PowerPoint, and Outlook, Copilot operates as an orchestration layer. Prometheus and Orchestrator pipelines ground queries with Microsoft Graph data before routing them to a foundation model. Historically, this has been OpenAI GPT-4 class models, including GPT-4 Turbo and GPT-4o. By late 2025, Microsoft confirmed that Claude Sonnet 4 and Claude Opus 4.1 are selectable in Copilot Studio, signaling a multi-model strategy for enterprise users who build custom Copilot agents.

While end-user Copilot in productivity apps typically routes to OpenAI models by default, the Studio environment allows builders to choose Anthropic models or even bring additional options through Azure AI Foundry. This ensures that Microsoft 365 Copilot can evolve from being OpenAI-centric into a broader orchestration platform.

·····

.....

Models available in GitHub Copilot.

GitHub Copilot is one of the most visible implementations of Microsoft’s model plurality. By 2025, the default for Copilot Chat shifted to GPT-4.1, with GPT-4o deprecated in this context. At the same time, GitHub integrated Claude Sonnet 4 and 4.5 into Pro, Business, and Enterprise tiers for more advanced reasoning. Documentation also lists Gemini 2.5 Pro as a model option in certain workflows, showing that GitHub Copilot spans multiple providers depending on client, IDE, and plan.

This approach reflects Microsoft’s strategy of allowing developers to leverage models best suited to their coding tasks, whether that means GPT-4.1 for broad tasks, Claude Sonnet for reasoning, or Gemini 2.5 Pro in specific scenarios.

·····

.....

Models behind Bing Copilot and Designer.

In consumer-facing services such as Bing Copilot and Designer, image and media generation rely on dedicated generative models. The default generator is DALL·E 3, with an option to use GPT-4o for image generation as well. This provides flexibility for creative tasks while ensuring that Copilot can serve both text-based and image-based use cases.

·····

.....

On-device models in Copilot+ PCs.

Windows Copilot+ PCs introduce a unique layer by embedding small models that run locally on the neural processing unit. The most prominent example is Phi Silica, a variant of Microsoft’s Phi-3.5 family, optimized for on-device tasks with a context window of roughly 4,000 tokens. These models handle latency-sensitive features and system functions without needing constant cloud calls, while larger tasks still route to cloud LLMs.

By combining cloud-based GPT, Claude, and Gemini models with local Phi Silica, Copilot+ PCs provide hybrid AI experiences tailored to device performance and privacy needs.

·····

.....

Copilot Studio and the bring-your-own-model preview.

Copilot Studio has become Microsoft’s most flexible environment for model access. In 2025, it introduced Claude Sonnet 4 and Opus 4.1 as selectable defaults for custom agents. It also began rolling out Bring Your Own Model (BYOM), allowing enterprises to integrate models from the Azure AI Foundry catalog. This catalog includes OpenAI GPT-4o and GPT-4.1, GPT-4o-mini, Microsoft’s Phi-3.5 family, and third-party models such as Meta Llama, Mistral, Cohere, and others.

Although BYOM is still in preview, it positions Copilot Studio as a multi-provider orchestration platform where organizations can align model selection with their workloads, governance requirements, and cost strategies.

·····

.....

Table — Microsoft Copilot models by surface.

Surface	Available models	Notes
Microsoft 365 Copilot	OpenAI GPT-4 class (GPT-4 Turbo, GPT-4o, GPT-4.1); Anthropic Claude Sonnet 4, Opus 4.1 (Studio)	Orchestration with Microsoft Graph grounding
Copilot Studio	GPT-4 family, Claude Sonnet 4/Opus 4.1, BYOM via Azure AI Foundry (Phi, Llama, Mistral, Cohere)	Model choice per agent; BYOM in preview
GitHub Copilot	GPT-4.1 (default), Claude Sonnet 4/4.5, Gemini 2.5 Pro	Model availability varies by IDE and subscription plan
Bing Copilot / Designer	DALL·E 3, GPT-4o for image generation	Dedicated for creative media tasks
Windows Copilot+ PCs	Phi Silica (on-device SLM)	Hybrid local + cloud architecture

This table consolidates the current landscape of models powering Copilot across its different platforms.

·····

.....

Operational recommendations.

For enterprise Microsoft 365 users, expect GPT-4 class models by default but explore Copilot Studio to deploy Claude or BYOM for specific agents. For developers, GitHub Copilot offers a broader palette: GPT-4.1 for coding, Claude Sonnet for reasoning, and Gemini 2.5 Pro for specialized contexts. For creative professionals, Bing Copilot and Designer provide DALL·E 3 and GPT-4o for image workflows. For device manufacturers and IT teams, Copilot+ PCs demonstrate how on-device Phi models can reduce latency and protect data while integrating with cloud models when needed.

By consolidating OpenAI, Anthropic, Google, and Microsoft’s own models under one orchestration strategy, Microsoft Copilot has become a flexible multi-model ecosystem rather than a single assistant. This diversity ensures that the platform can adapt to coding, productivity, creative, and enterprise governance scenarios with model selection tuned to each workload.

.....

FOLLOW US FOR MORE.

DATA STUDIOS

.....[datastudios.org]