Deepseek V4 Pro Dspark
DeepSeek V4 Pro Dspark is a large-scale Mixture-of-Experts (MoE) chat model built by DeepSeek, supporting a context length of one million tokens. It is designed for highly efficient long-context processing, utilizing a hybrid attention architecture to reduce inference FLOPs and KV cache requirements.
Input
Output
Context
1049K
Max Output
384K
Parameters
1650.5B
Technical Specifications
Model TypeChat
Context Window1,048,576 tokens
Max Output Tokens384,000 tokens
Parameters1650.5B
Release DateJun 27, 2026
Training CutoffNot available
Licensemit
Open SourceYes
Input Modalities
Text
Output Modalities
Text
Capabilities
Resources & Links
More Models by DeepSeek
Deepseek V3.1 Base131K context · 684.5B · Open weightsDeepseek Prover V1.5 SFT4K context · 6.9B · Open weightsDeepseek V3.2 Speciale164K context · 685.4B · Open weightsDeepseek Math 7B Base4K context · 7B · Open weightsDeepseek R1 Distill Qwen 32B131K context · 32.8B · Open weightsDeepseek V3 Base131K context · 684.5B · Open weights
Related Tools
Data sourced from official provider APIs and documentation
Last updated: Aug 8, 2026
Your AI stack shouldn't stand still.
Every month new models become cheaper, faster, and more capable. Inferbase ensures your application automatically benefits without changing a single API call.