DeepSeek: DeepSeek V4 Pro logo

MODELS

DeepSeek: DeepSeek V4 Pro

by DeepSeek

modellead-sourceopenrouter-modelslarge-language-modelmixture-of-expertslong-contextcodingreasoningdeepseeksource:api-docs.deepseek.comtool-callingjson-output
Visit WebsiteDocumentation

Overview

DeepSeek-V4-Pro is a DeepSeek Mixture-of-Experts language model with 1.6T total parameters, 49B activated parameters, and a 1M-token context window.

Details

DeepSeek-V4-Pro is part of the DeepSeek V4 Preview release. Supplied DeepSeek sources describe it as a text-modality Mixture-of-Experts model with 1.6T total parameters, 49B activated parameters, and a 1M context window. DeepSeek API documentation lists deepseek-v4-pro as an allowed chat-completion model value and documents API features including thinking mode, reasoning_effort high/max, JSON output, streaming, and function tool calls. DeepSeek’s model card names DeepSeek-V4-Pro and DeepSeek-V4-Flash, gives a release date of April 24, 2026, and states MIT-licensed weights with API terms for hosted use.

When to Use

Use for evaluating a DeepSeek long-context model where the supplied sources list a 1M-token context window. Use for coding reasoning or tool-using chat-completion workflows that need DeepSeek API support for function tool calls streaming and JSON output. Use when comparing large MoE models by total parameters activated parameters context length and availability of downloadable weights.

Getting Started

  1. Read the DeepSeek V4 Preview release note for the official announcement and model naming.
  2. Review the DeepSeek V4 model card for modality
  3. modes
  4. context length
  5. licensing
  6. API terms
  7. and safety or data notes.
  8. Check the DeepSeek Models & Pricing page for current token pricing and supported API features for deepseek-v4-pro.
  9. Use deepseek-v4-pro as the model value with the DeepSeek chat-completion API.
  10. For self-hosting or weight access
  11. review the official deepseek-ai/DeepSeek-V4-Pro Hugging Face model page.

Key Features

  • •Mixture-of-Experts architecture
  • •1.6T total parameters and 49B activated parameters
  • •1M-token context window listed in DeepSeek sources
  • •Chat-completion API access via model value deepseek-v4-pro
  • •Thinking modes including non-think
  • •think-high
  • •and think-max described in the model card
  • •JSON output
  • •streaming
  • •and function tool calls documented for the API

Capabilities

  • •long-context text generation
  • •advanced reasoning
  • •coding
  • •tool calling
  • •JSON output
  • •streaming chat completions

Last updated Jun 1, 2026