DeepSeek-V4-Pro

Description

DeepSeek-V4-Pro is a Mixture-of-Experts (MoE) language model with 1.6 trillion total parameters and 49 billion activated parameters. It features a hybrid attention architecture combining Compressed Sparse Attention (CSA) and Heavily Compressed Attention (HCA), achieving 27% of single-token inference FLOPs compared to DeepSeek-V3.2 at 1M-token context. Post-training uses a two-stage pipeline: independent domain-expert cultivation (SFT + GRPO) followed by unified model consolidation via on-policy distillation.

This container image is classified as a Pre-Release candidate (https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-software-license-agreement/)

Third-Party Community Consideration:

This model is not owned or developed by NVIDIA. This model has been developed and built to a third-party's requirements for this application and use case; see link to Non-NVIDIA DeepSeek-V4-Pro Model Card.

License and Terms of Use:

GOVERNING TERMS: The NIM container is governed by the NVIDIA Software License Agreement and Product-Specific Terms for NVIDIA AI Products. Use of this model is governed by the NVIDIA Open Model Agreement.

Additional Information: MIT License.

Deployment Geography:

Global

Use Case:

DeepSeek-V4-Pro is intended for text generation and agentic workflows such as:

• Advanced reasoning and mathematical problem-solving
• Code generation and software engineering assistance
• Agentic multi-step task execution and tool use
• Long-context document analysis and summarization (up to 1M tokens)
• General conversational assistant use

Release Date:

NGC 04/24/2026 via DeepSeek-V4-Pro on NGC HuggingFace 04/24/2026 via DeepSeek-V4-Pro

Reference(s):

• DeepSeek-V4-Pro on HuggingFace
• DeepSeek-V4 Technical Report
• DeepSeek Chat Interface
• DeepSeek Discord Community

Model Architecture:

Architecture Type: Transformer Network Architecture: Sparse-Attention Mixture of Experts (MoE) with hybrid attention (CSA + HCA) Total Parameters: 1.6T Active Parameters: 49B

Input:

Input Types: Text Input Formats: String Input Parameters: One Dimensional (1D) Other Input Properties: Supports multi-turn conversations with system prompts, user messages, and assistant responses. Maximum context length of 1,048,576 tokens (1M). Three reasoning modes: Non-think (fast), Think High (logical analysis), and Think Max (full reasoning extent).

Output:

Output Types: Text Output Format: String Output Parameters: One Dimensional (1D) Other Output Properties: Supports structured JSON output, function/tool calling, and reasoning content when enabled.

Our AI models are designed and/or optimized to run on NVIDIA GPU-accelerated systems. By leveraging NVIDIA's hardware (e.g. GPU cores) and software frameworks (e.g., CUDA libraries), the model achieves faster training and inference times compared to CPU-only solutions.

Software Integration:

Runtime Engines:

• SGLang

Supported Hardware:

• NVIDIA Blackwell (B200)

Operating Systems: Linux

The integration of foundation and fine-tuned models into AI systems requires additional testing using use-case-specific data to ensure safe and effective deployment. Following the V-model methodology, iterative testing and validation at both unit and system levels are essential to mitigate risks, meet technical and functional requirements, and ensure compliance with safety and ethical standards before deployment.

Model Version(s)

deepseek-ai/DeepSeek-V4-Pro

Training, Testing, and Evaluation Datasets:

Training Dataset

Data Modality: Text Text Training Data Size: More than 10 Trillion Tokens Training Data Collection: Undisclosed Training Labeling: Undisclosed Training Properties: Two-stage post-training pipeline: (1) independent cultivation of domain-specific experts via SFT and RL with GRPO, (2) unified model consolidation via on-policy distillation. Uses Muon optimizer for faster convergence and training stability.

Testing Dataset

Testing Data Collection: Undisclosed Testing Labeling: Undisclosed Testing Properties: Undisclosed

Evaluation Dataset

Evaluation Benchmark Score:

Evaluation Data Collection: [Automated] Evaluation Labeling: [Human] Evaluation Properties: Evaluated on competitive programming, mathematical reasoning, and general reasoning benchmarks.

Inference

Acceleration Engine: SGLang Test Hardware:

• NVIDIA B200 Precision formats: FP8 Mixed (MoE experts in FP4, other parameters in FP8)

Ethical Considerations

NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.

Users are responsible for model inputs and outputs. Users are responsible for ensuring safe integration of this model, including implementing guardrails as well as other safety mechanisms, prior to deployment.

Please report model quality, risk, security vulnerabilities or NVIDIA AI Concerns here.

Get Help

Getting started with the NIM

Deploying and integrating the NIM is straightforward thanks to our industry standard APIs. Visit the NIM Container page for release documentation, deployment guides and more.

NVIDIA Developer Community Forum

Get access to community knowledge base articles and support cases (NVIDIA Developer Forums).