Forum Discussion

SamLS42's avatar
SamLS42
Occasional Reader
Oct 28, 2025

Reasoning Effort for Foundry Agents

I am currently using the Azure AI Foundry Agents API and noticed that unlike the base completions endpoint, there is no option to specify the "Reasoning Effort" parameter. Could you please confirm if this feature is supported in the Agents API?

If not yet supported, are there any plans to introduce Reasoning Effort control for the Agents API in future releases?

agents
azure ai foundry

1 Reply

  • Surya_Narayana's avatar
    Surya_Narayana
    MCT
    Oct 30, 2025

    hi SamLS42​  This is a very relevant point when working with reasoning-capable models in Foundry.

    For the data-plane models under Azure AI Foundry Models (for example the “o*” series reasoning models) the reasoning_effort parameter is documented. You can set it to values like low, medium, or high.

    For example, the documentation says: “Reasoning models with reasoning_effort allow you to adjust the model’s cognitive load” (e.g., Grok 3 Mini model)

    Reference sites:

    https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/reasoning?

    https://ai.azure.com/catalog/models/grok-3-mini?

    I couldn’t locate any official documentation that confirms that the Agents API supports the reasoning_effort parameter explicitly. The reference for the Agents service and quick-start articles don't mention it.

    Therefore, the safe assumption is: No, you cannot currently set reasoning_effort via the Agents API endpoint — at least not publicly documented at this time.

    Next ?

    Microsoft’s public roadmap doesn’t show a guarantee of when (or whether) reasoning_effort will be exposed for the Agents API.

    You may want to open a support ticket or flag this feature request via the Azure feedback forums to see if it is under design.

    In the meantime: if you need control over reasoning depth, consider using the underlying model invocation (via Foundry Models or the OpenAI-style endpoint) where reasoning_effort is supported, then wrap that in your agent logic.

Resources