A Subtle Upgrade in DeepSeek Rivals the Introduction of a New Model
2025-03-26 / Read about 0 minute
Author:小编   

Last night, DeepSeek unveiled DeepSeek-V3-0324 on the Hugging Face platform, marking a significant milestone. This model boasts 685 billion parameters, closely mirroring its predecessor, V3, which featured 671 billion parameters, while also embracing the MoE architecture. Notably, DeepSeek-V3-0324 now supports the more permissive MIT open-source license. Despite the similar parameter count, DeepSeek-V3-0324 showcases enhanced reasoning and front-end development capabilities, harmonizing its writing style with the R1 model and undergoing a series of meticulous optimizations. This update underscores DeepSeek's relentless pursuit of innovation in AI model development and its steadfast commitment to open source principles.