DeepSeek released V4, an advanced AI model with a 1.6-trillion-parameter Pro version and a 1-million-token context window. The model leads open-source models in world-knowledge benchmarks and emphasizes drastically reduced training and inference costs.

Technical Specifications

  • Parameter count: 1.6 trillion (Pro version)
  • Context window: 1 million tokens
  • Focus: Cost efficiency in training and inference

Performance

The model leads open-source models in world-knowledge benchmarks. The release emphasizes accessibility through significantly reduced operational costs compared to competitors.

Strategic Position

DeepSeek V4 positions itself as a cost-effective alternative to premium models while maintaining strong benchmark performance, targeting developers and enterprises seeking high capability at lower cost.

Source: Bloomberg via AI Daily Digest