This topic describes service activation, endpoints, request parameters, response fields, supported features, and billing for SiliconFlow DeepSeek models.
Alibaba Cloud Model Studio offers DeepSeek model services from two inference service providers. The SiliconFlow provider supports a longer context, while the Alibaba Cloud Model Studio provider has more relaxed throttling limits and supports web search and context cache.
Activate the service
- Go to the Model Studio console, search for
deepseek, find the SiliconFlow DeepSeek model card, and click Activate Now. - In the dialog box, confirm the activation and authorization.
Prerequisites
Before calling a model, obtain an API Key and configure it as an environment variable. If you use an SDK, you must also install the SDK.
Service endpoints
This topic applies only to the China (Beijing) region. Replace {WorkspaceId} with your actual workspace ID.
| Call method | SDK base_url | HTTP request URL |
|---|---|---|
| OpenAI compatible | https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1 | POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/compatible-mode/v1/chat/completions |
| DashScope | https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1 | POST https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/text-generation/generation |
Call methods
The OpenAI-compatible API supports calls through the OpenAI Python SDK, Node.js SDK, or HTTP. enable_thinking is not a standard OpenAI parameter: pass it in extra_body for the Python SDK and as a top-level parameter for the Node.js SDK.
DashScope supports calls through the Python SDK, Java SDK, or HTTP. DashScope Java SDK version 2.19.4 or later is required.
Request parameters
| Parameter | Description |
|---|---|
model | The ID of the model to call. |
messages | The input message list. |
enable_thinking | Controls whether thinking mode is enabled for models that support this parameter. |
stream | Set to true to enable streaming output. |
stream_options.include_usage | Set to true to return token usage in a streaming response. |
incremental_output | For DashScope calls, set to true to enable incremental output. |
result_format | For DashScope calls, set to message to return results in message format. |
Response fields
| Field | Description |
|---|---|
reasoning_content | The complete thinking process. |
content | The complete response. |
usage | Token usage information. |
Other features
| Model | Multi-turn conversation | Function calling | Prefix completion | Web search | Context cache | Structured output |
|---|---|---|---|---|---|---|
| siliconflow/deepseek-v3.1-terminus | Supported | Supported | SupportedNon-thinking mode only | Not supported | Not supported | Not supported |
| siliconflow/deepseek-v3-0324 | Supported | Supported | Supported | Not supported | Not supported | Not supported |
Default parameter values
Model | Temperature | Top p | Repetition penalty | Presence penalty |
|---|---|---|---|---|
siliconflow/deepseek-v3.1-terminus | 1.0 | 1.0 | - | - |
siliconflow/deepseek-v3-0324 | 1.0 | 1.0 | - | - |
Models and billing
SiliconFlow uses its self-developed inference engine to provide low-latency, highly stable inference services for DeepSeek models.
- Hybrid Thinking Models (thinking is controlled by the
enable_thinkingparameter): siliconflow/deepseek-v3.1-terminus - Non-thinking models:
siliconflow/deepseek-v3-0324
In thinking mode, the chain of thought is billed as output tokens.