sophnet/DeepSeek-v3

DeepSeek-V3, launched by Hangzhou DeepSeek AI Fundamental Technology Research Co., Ltd., is a massive-scale language model featuring a Mixture-of-Experts (MoE) architecture. With a total of 671 billion parameters, it activates only about 37 billion during inference, significantly reducing computational demands and costs. The model excels in text understanding, coding, mathematics, and Chinese language processing, capable of handling long-form texts while maintaining logical coherence. Its performance rivals that of leading international closed-source models such as GPT-4o, while simultaneously driving the wider adoption of large-scale models with greater efficiency and lower costs.