On August 20, DeepSeek announced the open-sourcing of its new V3.1-Base model on Hugging Face. According to the company, the model has approximately 685 billion parameters and its context length has been extended to 128K.
Earlier that evening, DeepSeek had notified users that its online model had been upgraded to V3.1 with a 128K context length. This version can be tested on the official website, app, and mini-program, with the API call method remaining unchanged.
The company noted that there is no confirmed release date for the highly anticipated DeepSeek R2 model. [iThome, in Chinese]
Related