Core Viewpoint - DeepSeek has upgraded its online model to version V3.1, extending the context length to 128k and updating the user interface, which has generated significant attention in the AI community [1][3]. Summary by Sections Model Release - DeepSeek has released a new model, DeepSeek-V3.1-Base, on Hugging Face, which quickly rose to the 4th position on the platform's popular model list within hours of its launch [1]. - The model maintains similar parameters to its predecessor, DeepSeek-V3, and utilizes a mixture of experts (MoE) architecture [1]. Community Reaction - Reactions on social media regarding the update are mixed; some users express excitement, viewing it as a precursor to future models DeepSeek-V4 and DeepSeek-R2, while others feel the update does not meet the company's previous standards [3].
DeepSeek开源新基础模型，但不是V4，而是V3.1-Base
机器之心·2025-08-20 00:15