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.
[This article is prerelease documentation and is subject to change.]
Your agents can use a deep reasoning model to better understand and complete complex tasks. This capability enhances an agent's decision making and returns more accurate responses.
You can add deep reasoning models to agents with generative orchestration and deep reasoning turned on.
Note
Adding a deep reasoning model in Copilot Studio is currently only supported in some regions:
- In the United States, deep reasoning uses the Azure OpenAI o3 model
- In the EU (excluding the United Kingdom), deep reasoning uses the Azure OpenAI o1 model
Deep reasoning models in Copilot Studio are in preview, and do not make data residency commitments. Your data may be transferred to, and stored and processed in, the United States or any other country in which Microsoft or its subprocessors operate.
For organizations using pay-as-you-go billing: Although deep reasoning is only available in some regions, processing for pay-as-you-go messages may occur outside your agent's home region.
Once you turn on deep reasoning for your agent, the agent determines which tasks to apply a deep reasoning to, or you can add the reason keyword in the agent's instructions to make the agent apply deep reasoning for specific activity.
Important
This article contains Microsoft Copilot Studio preview documentation and is subject to change.
Preview features aren't meant for production use and may have restricted functionality. These features 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.
How does deep reasoning work?
Deep reasoning uses an advanced AI model to complete tasks requiring logical reasoning, problem solving, and step-by-step analysis. This model allows your agent to complete complex tasks with greater accuracy, though with slower response times, than default models used by Copilot Studio.
Your agent determines which tasks or steps benefit from applying a deep reasoning model. You can also tell your agents to use deep reasoning to complete tasks by using the keyword reason in the agent instructions.
Note
Use of deep reasoning consumes billed messages. For information on billed message rates for deep reasoning and other agent activities, see Billing rates and messages.
Turn on access to deep reasoning models
In order to use reasoning with an agent, you must turn on generative orchestration and access to deep reasoning models for that agent:
Open the Settings for your agent.
Turn on Deep reasoning (preview).
Use a deep reasoning model to complete tasks
Agents with access to deep reasoning models don't use them for all tasks or steps they perform. Instead, agents can use deep reasoning models to complete specific tasks.
Your agent determines which tasks or steps would benefit from a deep reasoning model.
You can also explicitly instruct the agent to use deep reasoning for specific tasks. Tell the agent to use reason in the agent instructions for completing a specific task or step.
For example, these instructions include the keyword reason in one of the steps. The agent uses the Azure OpenAI o3 model for that step only:
Parse the request body to identify: Location (e.g., Arizona) and services requested (e.g. cleaning, landscaping, maintenance).
Use Get suppliers from ERP flow to retrieve internal supplier data.
Search your internal KB for information on suppliers capacity, location, and performance.
Run search the web flow and extract Market Share, Revenue, Rank and other relevant information from all the suppliers.
Using the data, use reason to provide one top recommendation to the user request along with the relevant context.
Note
An agent can use deep reasoning models for multiple tasks or steps, but each one must contain the keyword reason. Because deep reasoning models take longer to respond, using models for multiple steps can substantially slow down your agent's response times.
Reviewing reasoning model behavior
You can trace when and how your agent uses deep reasoning models in the Activity page.
The activity map shows a deep reasoning node where the agent used a deep reasoning model.
To see details, select the reasoning node to expand it. The expanded node explains the reasoning steps and data the model used, and the result the model outputted.