DeepSeek-V3 model upgrade Users can experience it by turning off "Deep Thinking"

Zhitong
2025.03.25 03:30
portai
I'm PortAI, I can summarize articles.

DeepSeek launched a small version update of the DeepSeek-V3 model on the evening of March 24, increasing the parameters from 671 billion to 685 billion. This update is not the anticipated DeepSeek-V4 or R2; users can experience the new model by turning off the "Deep Thinking" option. The API interface and usage method of the DeepSeek V3 model remain unchanged and have been integrated by major companies such as Tencent, Baidu, Alibaba Cloud, and ByteDance. API top-ups have been reopened, with the calling price adjusted to 2 yuan per million input tokens and 8 yuan per million output tokens, still below the industry average