Call o3-pro in Azure OpenAI using C# SDK
Utilizing GPT-4 for software development accelerates project initialization and code generation, enabling developers to focus on complex tasks and improve overall efficiency.
Azure OpenAI supports now o3-pro model. This model is different from other ones, here is a brief summary of its advantages: The o3-pro model excels at advanced reasoning, delivers high accuracy, supports long-context tasks, and integrates powerful tools like web search and code execution. It’s cost-effective, reliable, and ideal for complex workflows in research, business, and creative fields.
You can find tons of example on how you can call Azure OpenAI models with C# Nuget Package but o3-pro is a slightly different beast. The usual problem is you will use existing code and you got this error.
| |
This error is due to the fact that o3-pro does not support chat model, you should use a response model where you create a question and then you got the answer (as well as some other details of the inner reasoning).
To solve this problem you first need to use the latest preview packate of Nuget, the 2.2.0-beta.5 at the time of this post.
<PackageReference Include="Azure.AI.OpenAI" Version="2.2.0-beta.5" />Then this is the C# code that you can use to call the model and examine the answer.
string endpoint = "https://yyyyyyyyy.openai.azure.com/"; // Your Azure endpoint
string apiKey = "xxxxxxxxxxxxx; // Your API Key
string deploymentName = "o3-pro"; // Deployment name for o3-pro
var options = new AzureOpenAIClientOptions(AzureOpenAIClientOptions.ServiceVersion.V2025_04_01_Preview);
AzureOpenAIClient azureClient = new(
new Uri(endpoint),
new AzureKeyCredential(apiKey),
options);
// Can't use this because Chat client is not supported by o3-pro
// it will throw The chatCompletion operation does not work with the specified model, o3-pro. Please choose different model and try again
//ChatClient chatClient = azureClient.GetChatClient(deploymentName);
//var message = ChatMessage.CreateUserMessage("Which is the capital of France?");
//var response = chatClient.CompleteChat(message);
#pragma warning disable OPENAI001 // Type is for evaluation purposes only and is subject to change or removal in future updates. Suppress this diagnostic to proceed.
var responseClient = azureClient.GetOpenAIResponseClient(deploymentName);
#pragma warning restore OPENAI001 // Type is for evaluation purposes only and is subject to change or removal in future updates. Suppress this diagnostic to proceed.
var response = responseClient.CreateResponse("Which are the real benefit of async in C#? What whould I lose if I do not use it?");
Console.WriteLine("Response:" + response);
var output = response.Value.OutputItems;
foreach (var item in output)
{
#pragma warning disable OPENAI001 // Type is for evaluation purposes only and is subject to change or removal in future updates. Suppress this diagnostic to proceed.
if (item is ReasoningResponseItem rri)
{
if (rri.SummaryParts.Count > 0)
{
Console.WriteLine("REASONING STEPS");
foreach (var sp in rri.SummaryParts)
{
Console.WriteLine($"ReasonStep: {sp}");
}
Console.WriteLine("");
}
}
else if (item is MessageResponseItem mri)
{
Console.WriteLine("ANSWER:");
foreach (var content in mri.Content)
{
Console.Write($"Content: {content.Text}");
}
}
#pragma warning restore OPENAI001 // Type is for evaluation purposes only and is subject to change or removal in future updates. Suppress this diagnostic to proceed.
}
Console.ReadKey();The code is really simple you just need to use a different mode, and you need to inspect the response, to distinguish between the response that contains reasoning part and the part that contains real answer text..
Gian Maria