.Net: New Feature: Expose GPT-5 Verbosity, ReasonEffort= Minimal parameters to SK

[rwjdk]

---

name: Feature request
about: Suggest an idea for this project

---

Since Chat GPT-5 was released, the OpenAI API has two new parameters that can control the AI Response:

1. The verbosity parameter (low/medium/high) that controls the length of the answer back
2. The ReasonEffort = Minimal (SK already has high/medium/low, but with GPT-5 a 'minimal' value has also been introduced that SK does not currently support (if you send the value, you get an exception))

It would be nice if SK could support these

[rwjdk]

#12989 resolves the Reason Effort = Minimal part

[tomw-twinthread]

if you are using GPT 5 hosted on Azure AI Foundry and the chat completions api, which semantic kernel uses, I've found that the verbosity parameter is ignored, and the default seems to be medium or high, resulting in twice as many output tokens being returned as gpt 4.1 and thus it takes twice as long to return the full response.

using open ai api works as expected.

I submitted a ticket to Azure Support, and they said they currently have no plans to support the verbosity parameter on Azure AI for chat completions. Instead, they recommended migrating to the Responses API, where it's supported.

[rwjdk]

@tomw-twinthread : Thank you for the info 👍. Guess we need to keep using prompts to keep the response size low

[jooooel]

Related to this, do you know if there are plants to support for the Response API in the Azure OpenAI nuget package? Seems like there is no way of using it with Semenatic Kernel otherwise?