Azure OpenAI o4-mini slow respond

[Jerrick_James_Decena]

Hello everyone, I have a question regarding the response of o4-mini. We tried prompting in Azure AI foundry playground, and we are using o4-mini. What I have noticed is even with simple questions like “What is the difference between power and authority”. The respond will took 2 minutes and it is just the chain of thoughts and not a complete response. Is there anything that i can do to make it respond faster? Thanks

[_j]

O4-mini has a parameter ‘reasoning_effort’, where you can trade the expenditure on reasoning generation to speed. The performance will be related to your particular model datacenter deployment of OpenAI services on Azure, something where this forum is not going to be a primary support source.

The endpoint returns a list of response items, so if you only get the first one parsed out, especially using the SDK’s helper response.output and just getting index:0, you are going to receive a reasoning summary and not content.

Async client.responses doesnt work?

A bit more code that I worked up for you that you can run, just as I have done. I’ve added a little output parser to make sure you get the “content” from list items that can also have empty reasoning summaries. If again producing errors, you’ll receive the openai version actually employed. import asyncio import openai async def get_response_content(response) -> str: """ ### Demonstrates that reasoning models have multi-item output ### Prints indices of response items containing te…