DeepSeek Models

The platform provides the DeepSeek-R1-32K and DeepSeek-V3-32K models.

DeepSeek-R1

DeepSeek-R1 671B, a full-scale version of the DeepSeek-R1 model, delivers on-par performance with OpenAI's o1 in a range of tasks, such as mathematics, coding, and natural language processing. DeepSeek-R1-32K-0.0.1 supports inference deployment.

The platform provides preconfigured inference services. You can call the inference services through the experience center or APIs.

The optimal model has been pre-built. There is no need to compress the model before deployment.

**Table 1** Inference features
Inference Precision	INT8
Sequence length	32K
Minimal inference unit	16
Concurrency	256

DeepSeek-V3

DeepSeek-V3 is a strong Mixture-of-Experts (MoE) model with 671B total parameters (with 37B activated for each token). It demonstrates remarkable performance across a wide range of tasks, such as long-context processing, code generation, mathematical reasoning, general knowledge tasks, and Chinese language understanding. DeepSeek-V3-32K-0.0.1 supports inference deployment.

The optimal model has been pre-built. There is no need to compress the model before deployment.