.Net: Bug: GPT-5 new reasoning level 'minimal' is not working in SK

[dmm-l-mediehus]

Describe the bug
I tried setting OpenAIPromptExecutionSettings value of 'ReasoningEffort' to the new OpenAI value 'minimal', and I get back The provided reasoning effort 'minimal' is not supported.

To Reproduce
See above.

Expected behavior
To work with 'minimal' reasoning effort with new gpt-5 model. Should also be added to the struct ChatReasoningEffortLevel to have Minimal.

Screenshots

Platform

• Language: C#
• Source: SK 1.61.0
• AI model: OpenAI: gpt-5
• IDE: VS
• OS: Win11

Additional context

[tomw-twinthread]

@copilot fix this!

[rogerbarreto]

This is a new feature introduced by OpenAI which was not available in the current version 2.2.0 we are depending on. This will be updated as soon we have all our dependencies updated to the latest OpenAI SDK.

As a work around for convenience of time, you can inject a HttpHandler and override the requestBody sent to the backend, with minimum setting.

[budgetdevv]

This is a new feature introduced by OpenAI which was not available in the current version 2.2.0 we are depending on. This will be updated as soon we have all our dependencies updated to the latest OpenAI SDK.

As a work around for convenience of time, you can inject a HttpHandler and override the requestBody sent to the backend, with minimum setting.

Do you happen to have example code for this?

[tomw-twinthread]

Here is how to intercept this..

var   _builder = Kernel.CreateBuilder();

   _builder.Services.AddTransient<Gpt5ReasoningEffortHandler>();
   _builder.Services.AddHttpClient("LLM_AzureOpenAI")
     .AddHttpMessageHandler<Gpt5ReasoningEffortHandler>();

var    httpClient = _builder.Services.BuildServiceProvider()
     .GetRequiredService<IHttpClientFactory>()
     .CreateClient("LLM_AzureOpenAI");


   // gpt 5
   _builder.AddAzureOpenAIChatCompletion("gpt-5", endpoint, apiKey, httpClient: httpClient);

 public class Gpt5ReasoningEffortHandler : DelegatingHandler
 {
   protected override async Task<HttpResponseMessage> SendAsync(HttpRequestMessage Request, CancellationToken CancellationToken)
   {
     if (Request.RequestUri != null && Request.RequestUri.AbsoluteUri.Contains("gpt-5", System.StringComparison.OrdinalIgnoreCase))
     {
       if (Request.Content != null)
       {
         var body = await Request.Content.ReadAsStringAsync(CancellationToken);
         if (!string.IsNullOrWhiteSpace(body))
         {
           var json = JObject.Parse(body);
           // Set reasoning_effort to minimal if not already set
           if (string.IsNullOrEmpty(json["reasoning_effort"]?.ToString()))
           {
             json["reasoning_effort"] = "minimal";
           }
           
           var newBody = json.ToString();
           Request.Content = new StringContent(newBody, Encoding.UTF8, "application/json");
         }
       }
     }
     return await base.SendAsync(Request, CancellationToken);
   }
 }