On August 21, DeepSeek mentioned in its official article "Officially releasing DeepSeek-V3.1" that DeepSeek-V3.1 uses the parameter accuracy of UE8M0FP8Scale. In addition, V3.1 has made major adjustments to the word segmenter and chat template, which are significantly different from DeepSeek-V3. DeepSeek’s official Weibo said in the pinned message,UE8M0FP8 is designed for the next generation of domestic chips that will be released soon.
According to the official DeepSeek public account on the 21st, DeepSeek-V3.1 was officially released. This upgrade includes the following major changes:
Hybrid inference architecture:A model supports both thinking and non-thinking modes;
Higher thinking efficiency:Compared with DeepSeek-R1-0528, DeepSeek-V3.1-Think can give answers in a shorter time;
Stronger Agent capabilities:Through Post-Training optimization, the performance of the new model in tool usage and agent tasks has been greatly improved.
The official App and web model have been simultaneously upgraded to DeepSeek-V3.1. Users can freely switch between thinking mode and non-thinking mode through the "Deep Thinking" button.
In terms of price adjustments, DeepSeek stated that starting from the early morning of September 6, 2025, Beijing time, the prices for calling the DeepSeek open platform API interface will be adjusted as follows: implement a new version of the price list (as shown in the figure below, see the pricing page for details); cancel night time discounts.
Before September 6, all API services will still be billed according to the original price policy, and you can continue to enjoy the current discounts.
Daily economic news comprehensive public information