DeepSeek V2 Open-Source Model Dominates Rankings: 236B Parameter MoE Architecture Crushes International Giants with Superior Cost-Performance
Chinese AI startup DeepSeek has released its latest open-source large language model DeepSeek V2, featuring a 236 billion parameter Mixture of Experts (MoE) architecture with inference costs only 1/30th of OpenAI's GPT-4o, rapidly topping Hugging Face's trending charts with over 80,000 interactions.