Import DeepSeek V4 Flash and DeepSeek V4 Pro in OCI Generative AI

The DeepSeek V4 Flash and DeepSeek V4 Pro models are now compatible for import in OCI Generative AI.

The DeepSeek V4 series consists of Mixture-of-Experts (MoE) text-to-text language models optimized for reasoning, coding, and agentic workflows. The models support context lengths of up to 1 million tokens and include architectural enhancements designed to improve long-context efficiency, training stability, and overall performance. DeepSeek V4 Pro has 1.6 trillion total parameters and activates 49 billion parameters during inference, while DeepSeek V4 Flash has 284 billion total parameters and activates 13 billion parameters during inference.

In OCI Generative AI, use the following details for these models:

Hugging Face model IDs for import:

• deepseek-ai/DeepSeek-V4-Flash
• deepseek-ai/DeepSeek-V4-Pro

Recommended dedicated AI cluster unit shapes:

• deepseek-ai/DeepSeek-V4-Flash: H100_X4
• deepseek-ai/DeepSeek-V4-Pro: H200_X8

For the complete list of models compatible for import, see Compatible Models for Import. For available hardware and deployment steps, see Managing Imported Models. For information about the service, see the Generative AI documentation.