GPT-5.4 Mini support

[AhmadAmr]

Issue: GPT-5.4 Mini Not Supporting Function Tools in Chat Completions

I tried to use the new gpt-5.4-mini model, but I encountered an error returned from the API.

API Response

Function tools with reasoning_effort are not supported for gpt-5.4-mini in /v1/chat/completions. Please use /v1/responses instead.

is there is a work around ?

[stephentoub]

Sounds like you're using OpenAI's GetChatClient. Try using GetResponsesClient.

[AhmadAmr]

I was previously using Aspire AppHost + AddOpenAIClient to register my chat clients:

Aspire AppHost

var openai = builder.AddConnectionString("openai");

API

var openai = builder.AddOpenAIClient("openai");

openai.AddKeyedChatClient("gpt5-agent", "gpt-5.4-mini")
      .UseRateLimiting()
      .UseFunctionInvocation()
      .UseOpenTelemetry(configure: c => c.EnableSensitiveData = true);

openai.AddKeyedChatClient("gpt4-agent", "gpt-4.1")
      .UseRateLimiting()
      .UseFunctionInvocation()
      .UseOpenTelemetry(configure: c => c.EnableSensitiveData = true);

openai.AddEmbeddingGenerator("text-embedding-3-large");

I switched to manually creating the client using GetOpenAIResponseClient

#pragma warning disable OPENAI001
builder.Services.AddKeyedSingleton<IChatClient>("gpt5-agent", (serviceProvider, _) =>
{
    var responseBackedChatClient = serviceProvider
        .GetRequiredService<OpenAIClient>()
        .GetOpenAIResponseClient("gpt-5.4-mini")
        .AsIChatClient();

    return new ChatClientBuilder(responseBackedChatClient)
        .UseRateLimiting()
        .UseFunctionInvocation()
        .UseOpenTelemetry(configure: c => c.EnableSensitiveData = true)
        .Build(serviceProvider);
});
    #pragma warning restore OPENAI001

it works but Is this the correct way?

Thank you in advance

[stephentoub]

@sebastienros / @eerhardt , what's the expectation here for Aspire around which protocol gets used with the Aspire OpenAI APIs?

[eerhardt]

We are really behind in the Aspire OpenAI integration. We need to update to the latest versions.

Right now we only have "easy methods" for AddChatClient and AddEmbeddingGenerator. Sounds like we should make one for GetOpenAIResponseClient?
I opened Add GetOpenAIResponseClient APIs (microsoft/aspire#15449) for this.