.Net: Bug: Memory leak when sending a byte array in a chat completion

[joe-sch-next]

Describe the bug
When sending a byte array to 'GetChatMessageContentsAsync' specifically targeting OpenAi Gpt 4.1. I can see the memory usage spike by 80x the original byte arrays size.

To Reproduce
Steps to reproduce the behavior:

1. Send a byte array chat completion to GPT 4.1
2. Monitor memory usage
3. See it spike by 80x the byte array size.

Expected behavior
I would expect some memory usage to increase, however I do not expect a 80x increase, that seems a bit much?

Screenshots

Here is me passing an image that is around 5MB, prior to uploading (right up to the semantic kernel line) memory usage was around 35MB. After calling the semantic kernel code, memory shoots up to 385MB.

My code... I run my calls in parallel, with there being 2 parallel calls executing at the moment. I have also tried to make them sequential with the same result

Platform

• Language: C#
• Source: Microsoft.SemanticKernel 1.49.0 - OpenAI 2.2.0-beta.4
• AI model: gpt-4.1 (version:2025-04-14)
• IDE: Rider
• OS: Windows/Linux

[RogerBarreto]

@joe-sch-next

The provided example is not a end-to-end reproduction scenario, I have no visibility on

1. How histories variable as initialized.
2. What is actually happening in the ProcessChatResponse(chatResponse, response);

Anyways I tried to do something on the lines of your description and wasn't able to reproduce the issue. With the Memory profile working as expected.

var configuration = new ConfigurationBuilder()
    .AddUserSecrets<Program>()
    .Build();

var content = new ImageContent(dataUri: "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAI0AAABKCAYAA...");

var history = new ChatHistory
{
    new ChatMessageContent(AuthorRole.User, "Hi"),
    new ChatMessageContent(AuthorRole.User, [content])
};
var settings = new AzureOpenAIPromptExecutionSettings { Temperature = 0.0f };
var completionService = new AzureOpenAIChatCompletionService(configuration["AzureOpenAI:DeploymentName"]!, configuration["AzureOpenAI:Endpoint"]!, configuration["AzureOpenAI:ApiKey"]!);

var task = completionService.GetChatMessageContentsAsync(
    history,
    settings
);

var responses = await Task.WhenAll([task, task]).ConfigureAwait(false);

foreach (var response in responses)
{
}

Profiler code.