Back to all articles
•Technology

DeepSeek Unveils Largest Open-Weight AI Model Amidst Controversy

View original source

DeepSeek, a Chinese AI lab, has launched two preview versions of its newest large language model, DeepSeek V4. This update includes two models: DeepSeek V4 Flash and V4 Pro, both utilizing a mixture-of-experts approach to lower inference costs. The key features include:

  • Model Size & Parameters:

    • The V4 Pro model boasts 1.6 trillion parameters with 49 billion active, surpassing Moonshot AI’s Kimi K 2.6 and other competitors.
    • V4 Flash, the smaller model, has 284 billion parameters with 13 billion active.

  • Performance & Efficiency:

    • Architectural improvements over the previous version (V3.2) have increased efficiency and performance.
    • On reasoning benchmarks, V4 models challenge leading open and closed models, claiming to outperform OpenAI’s GPT-5.2 and Gemini 3.0 Pro on certain tasks.

  • Benchmark Comparison:

    • Despite competitive reasoning benchmarking, V4 models slightly lag behind the latest models like OpenAI’s GPT-5.4 and Google’s Gemini 3.1 Pro in specific knowledge tests.

  • Cost Efficiency:

    • The models are priced competitively, with the V4 Flash model being the most affordable, undercutting various top-tier AI models.

  • Limited Scope:

    • Unlike some closed-source competitors, the V4 models only support text, lacking capabilities in audio, video, or image processing.

The release occurred shortly after the U.S. accused China of stealing AI technology, with DeepSeek being specifically named by companies like Anthropic and OpenAI for potentially copying AI models.