Select a primary AI model for your agent

In Microsoft Copilot Studio, the primary AI model you select for your agent powers generative orchestration. Choose a model based on how much your agent needs in response speed, output quality, reasoning, and cost.

Primary model selection is separate from the settings for deep reasoning (preview), generative responses (preview), and prompt builder models.

Model availability by region

Copilot Studio models vary by intended use, release status, and region. You can see each model's tags when you change your agent's AI model.

The following tables show selected models' regional availability. GA means generally available, and cross-geo indicates that data might be processed outside your region.

Standard harness availability

US Government availability

Model use categories

Models are optimized for different purposes. Your agent can perform better when you choose a model with the strengths that fit your agent's purpose. For example, an agent that makes complex decisions can benefit from a deep model, while an agent expected to talk about a wide range of topics could use a general model.

The following table describes the model use tags, their strengths, and considerations to keep in mind if you use the model.

Model release types

Model tags identify release types, defaults, and availability requirements. Use generally available models for production and experimental or preview models for exploration and testing.

• Experimental: Used for experimentation, and not intended for production use. Subject to preview terms, and can have limitations on availability and quality. See Limitations of experimental and preview models.
• Preview: Might eventually become a generally available model, but isn't intended for production use. Subject to preview terms, and can have limitations on availability and quality. See Limitations of experimental and preview models.
• Generally available (GA): Models without a release tag are generally available and intended for production use. Availability can vary by region.
• Default: The default model for all agents, and usually the best performing generally available model. The default model is periodically upgraded as new, more capable models become generally available. Agents also use the default model as a fallback if a selected model is turned off or unavailable.
• Retired: Models that are no longer supported. Some models offer an extension of up to 30 days after retirement. Check retirement dates and extension availability, and learn how to continue using a retired AI model.
• Cross-geo: Might require data processing and storage outside of your organization's geographical boundaries. Your admin can turn data movement across regions on or off.
• Early access environment: An early release cycle environment. These environments receive new updates and features first. Learn more in Early release cycle environments.

External models

You can also add external AI models from Anthropic, xAI, or Mistral to your agent. Learn more in Choose an external model as the primary AI model.

Model retirement dates and extension availability

Limitations of experimental and preview models

You can explore and test experimental and preview models, but don't use them for production:

• They might show variability in performance, response quality, latency, or message consumption, and might time out or be unavailable.

• Usage of published agents with an experimental or preview model is billed at the Copilot Studio billing rates.

Experimental and preview models are subject to preview terms. Use these models to evaluate capabilities before an official release and share feedback in the Copilot Studio community.

Change your agent's AI model

Your agent starts with a default model optimized for most scenarios. To change your agent's model:

1. Go to your agent's Overview page.

2. In the Model section, select your agent's primary model. You can switch between experimental and production models at any time.

Admin controls for AI model selection

Administrators control access to preview and experimental AI models with these environment settings:

• Preview and experimental AI models: Turn on this setting to let makers use preview and experimental models in the environment.

• Move data across regions: Turn on this setting to make experimental models available. Data processed within a preview or experimental model might be processed and stored outside your organization's geographical boundaries. The tenant administrator manages this setting in the Power Platform admin center.

Admin controls and requirements for external models

Admins control whether makers can add external models to agents. To grant access to external models, admins must complete the following actions:

• Turn on external models in Power Platform admin center for the environment or the environment group.

• Allow access to each external model provider in the Microsoft 365 admin center. Learn more in the Microsoft 365 admin center documentation:

  • Connect to Anthropic LLM

  • Connect to Mistral

  • Connect to xAI

Preview models and external models are two different sets that can overlap but aren't the same, and their settings are separate. For example:

• Admins can block external models but allow preview or experimental models. In this case, makers can't use external models but can use preview, experimental, and generally available internal models.

• Admins can block preview or experimental models but allow external models. In this case, makers can't use any preview or experimental models, but can use any generally available external and internal models.