Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
AI capabilities evolve rapidly, and each generative model brings distinct strengths, whether it's faster responses, higher-quality outputs, or improved cost efficiency. With Copilot Studio, you can choose the best model for your agent's orchestration by using a simple dropdown menu.
Want to try out cutting-edge models before they're production-ready? Access the latest experimental models to evaluate them early. However, they might have limited testing, availability, and functionality.
Note
- As of October 27, GPT‑4o is retired in Copilot Studio for agents using generative orchestration, except for GCC customers who will continue using GPT‑4o. The new default model is GPT‑4.1, which delivers improved performance, reliability, and consistency across experiences. GPT‑4o remains available until November 26, 2025 if you enable the Continue using retired models option.
- For latest model availability, check the list available in the models dropdown of your agent's Overview page.
Important
- Experimental models are available for exploration and testing but aren't recommended for production use. Review Limitations of experimental and preview models before choosing an experimental or preview model for your agent.
- Data processed within an experimental model might be processed and stored outside of your organization's geographical boundaries. For more information, see Admin controls for AI model selection.
- This article contains Copilot Studio documentation on model selection (including experimental model previews) and is subject to change.
Model tags
Copilot Studio offers different types of models. These models types are based on their intended use and availability.
You can see each model's tags in the list of models in Copilot Studio.
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.
| Tag | Description | Strengths | Latency | Cost | Reasoning depth |
|---|---|---|---|---|---|
| Deep | Optimized for deliberate, multistep reasoning and tool-supported workflows. | Complex analytics, multihop reasoning, policy and contract analysis, troubleshooting with multisystem steps, and synthesis of long documents with citations | Highest | Highest | Multistep, tool-rich |
| Auto | Optimized for coverage across mixed workloads; routes queries dynamically. | Helpdesk and employee agents with mixed intents, blending knowledge and actions, and tier‑0 customer support with unpredictable complexity | Variable | Variable | Adaptive per turn |
| General | Optimized for speed and cost on everyday chat and light grounding. | Drafting, rewriting, summarizing, and translation, FAQ-style grounded answers, and simple action automation | Lowest | Lowest | Shallow-to-moderate |
Model availability
Models have different stages of release. You can try out new, cutting-edge Experimental and Preview models, or choose a reliable, thoroughly tested Generally available model.
The following table describes the model availability tags.
| Tag | Description |
|---|---|
| Experimental | Used for experimentation, and not recommended for production use. Subject to preview terms, and can have limitations on availability and quality. See Limitations of experimental and preview models. |
| Preview | Will eventually become a generally available model, but currently not recommended for production use. Subject to preview terms, and can have limitations on availability and quality. See Limitations of experimental and preview models. |
| No tag | Generally available. You can use this model for scaled and production use. In most cases, generally available models have no limitations on availability and quality, but some might still have some limitations, like regional availability. |
| 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 | When a new model becomes the default model, the old default model is retired. You can still use the retired model for up to one month after retirement. For more information, see Continue using a retired AI model. |
External models
You can also add external AI models from Anthropic to your agent. For more information, go to Choose an external model as the primary AI model.
Limitations of experimental and preview models
You can explore and test experimental 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.
If you publish an agent with an experimental model and users use the agent, that use is billed at the established rates.
Feel free to experiment with these models to explore capabilities. However, be cautious about deploying them in production environments.
Experimental models are subject to preview terms. These models are available before an official release so that you can get early access and provide feedback. If you're building a production-ready agent, see Microsoft Copilot Studio overview.
Change your agent's AI model
Your agent starts with a default model optimized for most scenarios. To change your agent's model:
Go to your agent's Overview page.
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 can allow or disallow makers to add experimental AI models to agents by using the following settings:
Administrators can choose to allow or disallow preview and experimental models in an environment. To use these models, Preview and experimental AI models must be turned on for your environment.
Data processed within an experimental model might be processed and stored outside of your organization’s geographical boundaries. To make experimental models available, your environment must have the Move data across regions setting turned on. This is an environment-level setting managed in the Power Platform admin center by the tenant administrator.