OpenAI Delphi Client Update

sgcWebSockets 2026.4.0 introduces a major expansion of the OpenAI API integration, bringing full support for the new Responses API (the official replacement for the deprecated Assistants API), Audio Speech text-to-speech, Fine-Tuning Jobs management, the Batch API for asynchronous bulk processing, the Uploads API for large file handling, and modernized Chat Completions with tool calling and structured output support. This article covers every new method, its parameters, and practical Delphi code examples.

1. Responses API (Replaces Assistants)

The Responses API is the most significant addition in this release. It replaces the deprecated Assistants API with a streamlined, stateless interface exposed through the /responses endpoint. All methods are available on the TsgcHTTP_API_OpenAI component. Unlike Assistants, each Responses API call is a single round-trip that can include tool definitions, file search, web search, and structured output -- all in one request.

Methods

Delphi Example

var
  OpenAI: TsgcHTTP_API_OpenAI;
  vResponse: String;
begin
  OpenAI := TsgcHTTP_API_OpenAI.Create(nil);
  Try
    OpenAI.OpenAIOptions.ApiKey := 'sk-your-api-key';

    // Create a simple response
    vResponse := OpenAI._CreateResponse('gpt-4o', 'Explain quantum computing');
    ShowMessage(vResponse);

    // Retrieve a previously created response
    vResponse := OpenAI._RetrieveResponse('resp_abc123');
    ShowMessage(vResponse);

    // Delete a stored response
    OpenAI._DeleteResponse('resp_abc123');
  Finally
    OpenAI.Free;
  end;
end;

2. Audio Speech (Text-to-Speech)

The Audio Speech API provides text-to-speech capabilities using OpenAI's TTS models. It supports two model tiers: tts-1 for low-latency streaming use cases and tts-1-hd for higher quality output. Six built-in voices are available: alloy, echo, fable, onyx, nova, and shimmer. The output can be returned in mp3, opus, aac, flac, wav, or pcm format.

Methods

Delphi Example

var
  OpenAI: TsgcHTTP_API_OpenAI;
  vAudioStream: TMemoryStream;
begin
  OpenAI := TsgcHTTP_API_OpenAI.Create(nil);
  Try
    OpenAI.OpenAIOptions.ApiKey := 'sk-your-api-key';

    // Generate speech using the 'alloy' voice
    oStream := TFileStream.Create('stream.mpeg', fmCreate or fmOpenRead);
    Try
      OpenAI._CreateSpeech('tts-1', 'Hello world', 'alloy', oStream);
    Finally
      oStream.Free;
    End;

    // Generate high-definition speech with 'nova' voice
    oStream := TFileStream.Create('stream.mpeg', fmCreate or fmOpenRead);
    Try
      OpenAI._CreateSpeech('tts-1-hd', 'Welcome to sgcWebSockets.', 'nova', oStream);
    Finally
      oStream.Free;
    End;
  Finally
    OpenAI.Free;
  end;
end;

3. Fine-Tuning Jobs

The Fine-Tuning Jobs API replaces the deprecated /fine-tunes endpoint with the new /fine_tuning/jobs endpoint. It provides full lifecycle management for fine-tuning operations: creating jobs, listing active and completed jobs, retrieving details, cancelling in-progress jobs, and streaming training events. This API supports fine-tuning of models like gpt-4o-mini-2024-07-18 using your own training data.

Methods

Delphi Example

var
  OpenAI: TsgcHTTP_API_OpenAI;
  vResponse: String;
begin
  OpenAI := TsgcHTTP_API_OpenAI.Create(nil);
  Try
    OpenAI.OpenAIOptions.ApiKey := 'sk-your-api-key';

    // Create a fine-tuning job
    vResponse := OpenAI._CreateFineTuningJob('gpt-4o-mini-2024-07-18', 'file-abc123');
    ShowMessage(vResponse);

    // List all fine-tuning jobs
    vResponse := OpenAI._ListFineTuningJobs;
    ShowMessage(vResponse);

    // Retrieve a specific job
    vResponse := OpenAI._RetrieveFineTuningJob('ftjob-xyz789');
    ShowMessage(vResponse);

    // List events for a job
    vResponse := OpenAI._ListFineTuningJobEvents('ftjob-xyz789');
    ShowMessage(vResponse);

    // Cancel an in-progress job
    OpenAI._CancelFineTuningJob('ftjob-xyz789');
  Finally
    OpenAI.Free;
  end;
end;

4. Chat Completions Updates

The Chat Completions API in sgcWebSockets 2026.4.0 has been modernized with several new request properties and response fields. These additions bring full support for tool/function calling, structured JSON output, deterministic generation via seeds, and parallel tool execution.

New Request Properties

New Response Fields

Delphi Example

var
  OpenAI: TsgcHTTP_API_OpenAI;
  vResponse: String;
begin
  OpenAI := TsgcHTTP_API_OpenAI.Create(nil);
  Try
    OpenAI.OpenAIOptions.ApiKey := 'sk-your-api-key';

    // Configure Chat Completions with new properties
    OpenAI.ChatCompletions.Model := 'gpt-4o';
    OpenAI.ChatCompletions.MaxCompletionTokens := 1024;
    OpenAI.ChatCompletions.Seed := 42;
    OpenAI.ChatCompletions.ParallelToolCalls := True;
    OpenAI.ChatCompletions.ResponseFormat := 'json_object';

    // Add a user message and create the completion
    OpenAI.ChatCompletions.AddMessage('user', 'List 3 benefits of Delphi in JSON format.');
    vResponse := OpenAI._CreateChatCompletion;
    ShowMessage(vResponse);
  Finally
    OpenAI.Free;
  end;
end;

5. Batch API

The Batch API allows you to send large groups of API requests for asynchronous processing. This is ideal for workloads that do not require immediate responses, such as bulk classification, embedding generation, or large-scale content moderation. Batch requests typically complete within 24 hours and offer a 50% cost reduction compared to synchronous API calls. All batch methods are available on the TsgcHTTP_API_OpenAI component through the /batches endpoint.

Methods

Delphi Example

var
  OpenAI: TsgcHTTP_API_OpenAI;
  vResponse: String;
begin
  OpenAI := TsgcHTTP_API_OpenAI.Create(nil);
  Try
    OpenAI.OpenAIOptions.ApiKey := 'sk-your-api-key';

    // Create a batch job targeting chat completions
    vResponse := OpenAI._CreateBatch('file-abc123', '/v1/chat/completions');
    ShowMessage(vResponse);

    // Check batch status
    vResponse := OpenAI._RetrieveBatch('batch_xyz789');
    ShowMessage(vResponse);

    // List all batches
    vResponse := OpenAI._ListBatches;
    ShowMessage(vResponse);

    // Cancel a batch if needed
    OpenAI._CancelBatch('batch_xyz789');
  Finally
    OpenAI.Free;
  end;
end;

6. Uploads API

The Uploads API enables uploading large files in multiple parts, which is essential when working with files that exceed the single-request upload limit (typically 512 MB). The workflow involves creating an upload session, adding parts sequentially, and then completing the upload to receive a File object that can be used with other API endpoints. All methods are available on the TsgcHTTP_API_OpenAI component through the /uploads endpoint.

Methods

Delphi Example

var
  OpenAI: TsgcHTTP_API_OpenAI;
  vUploadResponse: String;
  vPartResponse: String;
  vCompleteResponse: String;
begin
  OpenAI := TsgcHTTP_API_OpenAI.Create(nil);
  Try
    OpenAI.OpenAIOptions.ApiKey := 'sk-your-api-key';

    // Step 1: Create an upload session
    vUploadResponse := OpenAI._CreateUpload(
      'training_data.jsonl', 'fine-tune', 104857600, 'application/jsonl');
    ShowMessage(vUploadResponse);

    // Step 2: Add file parts
    vPartResponse := OpenAI._AddUploadPart('upload_abc123', 'C:\data\part1.jsonl');
    ShowMessage(vPartResponse);

    vPartResponse := OpenAI._AddUploadPart('upload_abc123', 'C:\data\part2.jsonl');
    ShowMessage(vPartResponse);

    // Step 3: Complete the upload
    vCompleteResponse := OpenAI._CompleteUpload('upload_abc123',
      '["part_def456", "part_ghi789"]');
    ShowMessage(vCompleteResponse);
  Finally
    OpenAI.Free;
  end;
end;