Recently, the DeepSeekR1 model has been upgraded. This is the first update since the model was launched in January 2025. What capabilities of the large model have been enhanced after the upgrade? Have downstream application scenarios become more enriched? Prof. Hongwei Wang, Associate Dean of Tongji-SEM, was interviewed by CCTV Financial News on the topic of the new DeepSeekR1 model driving industrial development. Below is the interview video and excerpts from the content.

A reporter from a futures company in Shanghai learned that with the upgrade of the DeepSeekR1 model, the response speed of the original large model application scenarios has been significantly improved.

According to a person from the futures company: “In the field of quantitative research, we conducted a modeling test. This complete test included the generation of codes for customer preprocessing, drawdowns, and so on. During this process, we found that its speed increased by nearly 60%, which is a qualitative leap compared to before.”

Recently, the company launched AI investment research services on its APP, which condenses and summarizes research reports. The company’s chief analyst said: “Currently, there are 148 futures and options varieties and more than 70 industries that can be viewed with one click and in a visual manner. It presents some trend views, directions, and policy impact analyses visually in the final conclusion for investors to refer to.”

Several third-party platforms have already connected to the new version of the DeepSeekR1 model. Industry insiders said that test results show that the new version of the DeepSeekR1 model can effectively provide more accurate and reliable results.

A product manager from a cloud computing service provider analyzed that: “Its programming ability has improved. We tested it with the Minesweeper game and found that the new version of the model generates more grids and the interface design is more in line with users’ daily usage habits. It has a stronger long-term thinking ability, and the new architecture model can think continuously for 20 minutes.”

Industry insiders said that the version upgrade of the DeepSeekR1 model is expected to drive the further development of the entire domestic large model industry chain. From upstream computing power chips, midstream language databases, to downstream deployment and construction of domestic large models, new opportunities for coordinated development will emerge.

Prof. Hongwei Wang, Associate Dean of Tongji-SEM, said: “The new version of the DeepSeekR1 model has reduced the difficulty of model development, allowing more enterprises to participate in the subsequent development and utilization of the model. We hope to form a closed-loop ecosystem that drives chip development with the model, promotes data construction with chips, and ultimately strengthens security, enhancing China’s say in the industry.”