Can't leverage JSON file as UserChatMessage: Expected a file with an application/pdf MIME type, but got unsupported MIME type 'application/json'

[justintoth]

Service

OpenAI

Describe the bug

I'm attempting to upload a JSON file and then use that within my user message prompts. However, after I upload the JSON file and reference its fileId in a UserChatMessage, it throws this exception:

System.ClientModel.ClientResultException: HTTP 400 (invalid_request_error: invalid_value)
Parameter: file_id
Invalid file data: 'file_id'. Expected a file with an application/pdf MIME type, but got unsupported MIME type 'application/json'.

It's confusing because clearly JSON files are supported (or else it wouldn't allow me to upload the file), and yet it's expecting a PDF for some reason. What am I missing?

public async Task TestJsonUpload()
{
	var openAiApiKey = RprConfigurationManager.GetAppSetting<string>("GenAiOpenAiApiKey");
	ChatClient client = new(model: "gpt-4.1", apiKey: openAiApiKey);
	OpenAIClient openAIClient = new(openAiApiKey);

	var json = """{"propertyId":37037566,"propertyMode":"Residential","propertyType":"Single Family","address":"10900 Fox Hedge Rd, Matthews, NC 28105","zip":"28105","centroid":{"latitude":35.0519990000,"longitude":-80.7527680000},"transactionType":"For Sale","listingStatus":"Closed","zoning":"R-3","listDate":"2024-11-07T00:00:00Z","offMarketDate":"2025-02-20T00:00:00Z","pendingDate":"2025-01-25T00:00:00Z","yearBuilt":1999,"bedrooms":5,"bathrooms":3,"livingArea":"3,760 sq ft","homeownersAssociationFeatures":{"Dues":"$1,175","Payments Per Year":"1"}}{"propertyId":37037566,"propertyMode":"Residential","propertyType":"Single Family","address":"10900 Fox Hedge Rd, Matthews, NC 28105","zip":"28105","centroid":{"latitude":35.0519990000,"longitude":-80.7527680000},"transactionType":"For Sale","listingStatus":"Closed","zoning":"R-3","listDate":"2024-11-07T00:00:00Z","offMarketDate":"2025-02-20T00:00:00Z","pendingDate":"2025-01-25T00:00:00Z","yearBuilt":1999,"bedrooms":5,"bathrooms":3,"livingArea":"3,760 sq ft","homeownersAssociationFeatures":{"Dues":"$1,175","Payments Per Year":"1"}}""";
	string? fileId = null;
	var fileClient = openAIClient.GetOpenAIFileClient();
	using (var stream = BinaryData.FromBytes(Encoding.ASCII.GetBytes(json)).ToStream())
	{
		OpenAIFile file = await fileClient.UploadFileAsync(
			stream,
			$"subject-property.json",
			FileUploadPurpose.UserData);
		fileId = file.Id;
	}

	var completionOptions = new ChatCompletionOptions
	{
		MaxOutputTokenCount = 350,
		Temperature = .4f,
		TopP = .8f
	};
	var chatMessages = new List<ChatMessage> { new UserChatMessage([ChatMessageContentPart.CreateFilePart(fileId)]) };

	ChatCompletion completion = await client.CompleteChatAsync(chatMessages, completionOptions);
	var message = completion.Content.Any() ? completion.Content[0].Text : null;
}

Steps to reproduce

Run the sample code, inserting your own API key.

Code snippets

OS

Windows 11 Preview for ARM

.NET version

9

Library version

2.2.0-beta.4

[justintoth]

I switched to using SystemChatMessage instead of UserChatMessage and the MIME type exception went away, however the AI model still isn't recognizing the uploaded file.

I don’t see any file uploaded in this conversation yet, including a property.json file. If you’d like me to review or analyze a file, please upload it, and I’ll be happy to help!

[TestMethod]
public async Task TestJsonUpload1()
{
	var openAiApiKey = RprConfigurationManager.GetAppSetting<string>("GenAiOpenAiApiKey");
	ChatClient client = new(model: "gpt-4.1", apiKey: openAiApiKey);

	var json = """
		   {
		    "propertyId": 37037566,
		    "propertyMode": "Residential",
		    "propertyType": "Single Family",
		    "address": "10900 Fox Hedge Rd, Matthews, NC 28105",
		    "zip": "28105",
		    "centroid": {
		        "latitude": 35.051999,
		        "longitude": -80.752768
		    },
		    "transactionType": "For Sale",
		    "listingStatus": "Closed",
		    "zoning": "R-3",
		    "listDate": "2024-11-07T00:00:00Z",
		    "offMarketDate": "2025-02-20T00:00:00Z",
		    "pendingDate": "2025-01-25T00:00:00Z",
		    "yearBuilt": 1999,
		    "bedrooms": 5,
		    "bathrooms": 3,
		    "livingArea": "3,760 sq ft",
		    "homeownersAssociationFeatures": {
		        "Dues": "$1,175",
		        "Payments Per Year": "1"
		    }
		}
		""";
	
	var completionOptions = new ChatCompletionOptions
	{
		MaxOutputTokenCount = 350,
		Temperature = .4f,
		TopP = .8f
	};
	var chatMessages = new List<ChatMessage>
	{
		new SystemChatMessage([ChatMessageContentPart.CreateFilePart(BinaryData.FromString(json), "application/json", "property.json")]),
		new SystemChatMessage("Do you have access to the property.json message that I uploaded?"),
	};

	ChatCompletion completion = await client.CompleteChatAsync(chatMessages, completionOptions);
	var message = completion.Content.Any() ? completion.Content[0].Text : null;
	Console.WriteLine(message);

[joseharriaga]

Thank you for reaching out, @justintoth ! At the moment, file inputs can only be PDF files. You can find more information here:
🔗 https://platform.openai.com/docs/guides/pdf-files?api-mode=chat

[justintoth]

@joseharriaga Gotcha... So my only options are:

1. To pass the JSON into the prompt as plain text. In which case, if I need to access it in three separate requests, I'll need to pass it in to all three requests separately.
2. To pass the JSON into the prompt as plain text, but use the Assistants API so that it remembers the JSON on the second and third requests.

Is this correct?

[joseharriaga]

@justintoth : You're correct. But there is also a third option: If you grab the latest preview of the .NET library, you can also use the Responses API. The Responses API can be stateless (like Chat) or stateful (like Assistants). To continue a conversation based on an existing Response, you just need to pass any new messages along with the PreviousResponseId in ResponseCreationOptions. Also, note that the current plan is to sunset the Assistants API in 2026, so if you're building entirely new code today, it's probably better to start with the Responses API instead.