Choosing an AI model for the agent
Choosing the right AI model helps your agent understand user intent, reason through tasks, and generate reliable responses. Your choice directly affects accuracy, performance, and cost.
Before you begin
Before you select a model:
-
Identify the use case that your agent must support, including accuracy, latency, and cost expectations.
-
Review Considerations for choosing an AI model to understand how model characteristics affect performance.
-
Check Available AI models to confirm which models are supported in your region and cloud provider.
-
Make sure that you have an existing agent or are ready to create one.
-
Verify that model selection is enabled for your tenant. Administrators can control this setting. For more information, see Managing model selection settings.
Assigning a model
If model selection is disabled by your administrator, the AI model dropdown is disabled and shows the currently selected model. You cannot change the model unless it is deprecated or unavailable. Contact your administrator if you need to select a different model. For more information, see Managing model selection settings.
To assign an AI model to your agent:
-
Create a new agent or open an existing one.
-
Locate the AI model drop-down list in the agent settings.
-
Review the available models and choose the one that matches your task needs.
-
Consider task complexity, input type, and performance requirements.
-
-
If more than five models are available, click View all foundation models to see the full list.
-
In the Model selection section, search for the model you need.
-
Hover over a model name to view details.
-
-
Select the model and click Save.
After selecting a model, continue configuring your agent by customizing prompts or adding knowledge.
Model availability and status
The AI model drop-down list includes models that are:
-
Provided by watsonx Orchestrate by default
-
Recommended for your region
-
Added through Agent Development Kit (ADK)
-
Connected through AI gateway
Model status tags
Some models display a status tag to indicate availability or lifecycle state.
|
Status |
Description |
Action required |
|---|---|---|
|
Default |
Model used automatically when no model is selected. |
No action is needed unless you want to change the model. |
|
Deprecated |
Model is supported now but might be withdrawn in a future release. |
Plan to migrate to another supported model. |
|
Not available |
The model is either no longer available for use or not accessible in the tenant. |
Select another supported model before deployment. |
|
Premier |
This is a premier model. Selecting it incurs additional MAUs beyond your standard included usage. |
Before you can select this model, your admin must enable premier models for your tenant. See Managing premier models. |
-
When you use a third-party model, the model is provided under its own license and is not governed by IBM. Review the license terms before use. Read the terms.
-
If you select a model with the Not available tag, the Pre-deployment summary displays an error. Choose another supported model to deploy the agent.
Order of models in the list
The AI model drop-down list displays models in the following order:
-
The currently selected model.
-
The default model, if it is not already selected.
-
Models that are listed as recommended in the ADK (if it is available in the data-center).
-
External models added through the AI Gateway, listed in alphabetical order.
-
Model policies configured through the ADK. For more information, see Configuring model policies.
What to do next
-
Customize prompts to tailor your agent’s behavior, see Customizing the welcome message and starter prompts.
-
Add knowledge sources to improve accuracy and domain coverage, see Adding knowledge to agents.
-
If you need additional models, configure them by using the ADK, see Adding custom AI models.