Universal.Alibaba.DashScope.Client 1.1.0

Universal.Alibaba.DashScope.Client

A .NET Standard 2.0 client for native DashScope generation, embeddings, reranking, applications, media, audio, tasks, files, fine-tuning, deployments, assistants, and threads. OpenAI- and Anthropic-compatible endpoints are also available.

Quick start

using Universal.Alibaba.DashScope.Client.V1;
using Universal.Alibaba.DashScope.Client.V1.Generation;

using var client = new DashScopeClient("your-api-key");

GenerationResponse response = await client.Generation.CreateAsync(new GenerationRequest
{
    Model = DashScopeModels.Text.QwenPlus,
    Input = new GenerationInput
    {
        Messages = new[] { GenerationMessage.User("Hello") }
    }
});

Console.WriteLine(response.Output.Choices[0].Message.Content);

DashScopeClient uses the international native endpoint by default. Configure another region, a workspace, timeout, or HTTP handler with DashScopeClientOptions.

using var client = new DashScopeClient(new DashScopeClientOptions
{
    ApiKey = "your-api-key",
    Endpoint = DashScopeEndpoints.UnitedStates,
    WorkspaceId = "your-workspace-id",
    Timeout = TimeSpan.FromMinutes(2)
});

Stream native text generation with client.Generation.StreamAsync, or call unified multimodal and Qwen-VL/audio models with client.MultimodalGeneration.StreamAsync. Multimodal messages preserve reasoning_content, fragmented tool_calls, and tool_call_id so agent histories can be replayed correctly.

Model identifiers are grouped under DashScopeModels by capability. OpenAICompatibleClient exposes both Chat Completions and Responses; AnthropicCompatibleClient exposes Messages. Compatibility base URIs can be derived from any native regional or workspace URI with DashScopeEndpoints.OpenAICompatible and DashScopeEndpoints.AnthropicCompatible.

Version 1.1

Universal.Alibaba.DashScope.Client.Compatible.OpenAI.Chat.Message.Content is now object rather than string. Existing text assignments and message constructors continue to work. Consumers that read response content should cast or convert it to string; request messages may now instead supply a JArray containing OpenAI-compatible text, image, audio, or video parts.

Version 1.1 also adds OpenAICompatibleClient.Responses, preserves reasoning and fragmented tool calls on native multimodal streams, completes the Anthropic-compatible thinking/error shapes, and includes qwen3.8-flash in the model catalogue.