It is learned that on the evening of May 29, DeepSeek announced that its R1 model has successfully completed an important version iteration, and the new version is named DeepSeek-R1-0528.

Improve thinking depth and reasoning ability
In the evaluation of the new version, DeepSeek-R1-0528 has achieved outstanding results in mathematics, programming, and general logic. It not only ranks first among domestic models, but also has approached the international top models such as o3 and Gemini-2.5-Pro in overall performance.
In addition to the improvement of reasoning ability, the new version of DeepSeek R1 has also been optimized in other aspects. In response to the “hallucination” problem, the new version of the model has been targeted. Compared with the old version, the updated model has reduced the hallucination rate by about 45% to 50% in rewriting and polishing, summarizing, and reading comprehension, so as to provide users with more accurate and reliable results.
DeepSeek-R1-0528 is still based on the DeepSeek V3 Base model launched in December 2024, but the computing power investment has been increased in the post-training stage, which significantly enhances the model’s thinking depth and reasoning ability. This achievement marks a major step forward in DeepSeek’s technical strength in the field of artificial intelligence.
Accelerate the development of the industry ecosystem
It has to be said that the application of large models has continued to explode since this year. Since the beginning of 2023, ChatGPT has detonated large model technology, and a “hundred-model war” has been set off at home and abroad. Now, after more than two years of technical drills, the “hundred-model war” has been fully upgraded to “a hundred schools of thought.”
In terms of the market, it is reported that Qwen3 is the latest generation of large-scale language model series of Alibaba (BABA) Tongyi Qianwen team, covering dense models and mixed expert (MoE) models, with significant improvements in reasoning, instruction following, intelligent agent capabilities and multi-language support.
It is worth mentioning that the Beijing Municipal Bureau of Economy and Information Technology recently issued the “Beijing Artificial Intelligence Empowerment New Industrialization Action Plan (2025)”, which proposed to strengthen model application financial services. Make good use of relevant funds such as municipal artificial intelligence, robots, advanced manufacturing and intelligent equipment to cultivate high-quality manufacturing enterprises, large models and intelligent innovation enterprises.
WiMi continues to explore the field of AI and large models
According to the data, Wimi Hologram Cloud Inc (WIMI), as an important innovator in the field of AI, has accelerated the exploration of AI and large model technologies through multi-dimensional layout in recent years, and its strategy covers multiple levels such as technology research and development, open source ecology, computing power infrastructure and industry application landing. In this regard, the market is generally optimistic about WiMi’s competitive advantage in the AI track, especially its continued outstanding achievements in AI business.
In fact, WiMi has built a “holographic cloud” platform covering the cloud and edge, open model code and computing power interface, supporting developers to call general large models such as DeepSeek for secondary development, and accelerating commercial verification. At the same time, its self-developed multimodal AI system integrates text, images, videos and other data to improve cross-modal understanding capabilities, and is applied to scenarios such as video generation and smart terminals.
It can be said that WiMi has made forward-looking layout of AI technology, and through the “integration” of industrial innovation, technological innovation and scenario innovation, it has focused on the artificial intelligence and large model industries, and has made intensive efforts and in-depth cultivation. It has adhered to the engineering path of technical research, software and hardware collaboration, application traction, and ecological cultivation to create a distinctive industrial ecology. In the future, it will also further optimize the vertical industry layout, and carry out more industrial practices in vertical fields such as automobiles, finance, and education.
Conclusion
Industry insiders pointed out that in 2025, AI will focus on commercial scenarios, and this year, the penetration of AI intelligent bodies in vertical industries has accelerated, and large models are leaping from “auxiliary tools” to “core productivity.” “Looking to the future, all companies need to actively help the development of the artificial intelligence industry, “accelerate the innovation of new quality productivity, promote the application and promotion of new technologies, new products, and new processes, and iterative upgrades, and promote more companies to ride on the “cloud” and develop with “AI”.




