It is learned that on March 25, DeepSeek, a Chinese artificial intelligence startup, announced the upgrade of V3 version, and the actual performance far exceeded expectations. The current version number is DeepSeek-V3-0324. After users log in to the official website, APP, and applet to enter the dialogue interface, they can experience it by closing deep thinking.
V3 model is coming, and all capabilities are fully upgraded
It is understood that the performance of DeepSeek-V3 model reasoning tasks has improved. The new version of the model draws on the reinforcement learning technology used in the training process of DeepSeek-R1 model, greatly improving the performance level in reasoning tasks, and has achieved a score higher than GPT-4.5 on the relevant evaluation sets of mathematics and code.
In fact, DeepSeek originally planned to release R2 in early May this year, but the specific time has not yet been disclosed. Today, the outside world’s attention to DeepSeek-V3-0324 is not only because of the improvement of the capabilities of this version, but also because it is speculated that the release of DeepSeek’s newer generation of V4 and R2 large models is not far away.
In the context of OpenAI’s GPT big model to integrate the general big model and the reasoning big model, the outside world is paying attention to DeepSeek and hopes that the new model will further improve its performance in code generation and multi-language reasoning.
Tencent is the first to access the latest version of DeepSeek V3
At the same time, Tencent (TCEHY) announced a major upgrade and accessed two models at the same time: the official version of the deep thinking model “Tencent Hunyuan T1” and the latest version of DeepSeek V3-0324. Tencent said that it was one of the first AI applications to access the DeepSeek V3-0324 version, and it took only one day from model open source to online.
Recently, Tencent Yuanbao has continued to update at a high frequency, iterating 30 versions in 35 days, and continuously launching practical functions. All capabilities are free and open, unlimited times, covering more work, study and life scenarios. Now, Tencent Yuanbao’s “Hunyuan + DeepSeek” dual models both support deep thinking and give answers in seconds.
At present, various industries are accelerating to embrace AI, and DeepSeek’s artificial intelligence model represents “innovation that promotes efficiency.” At the end of January this year, the Deepseek app topped the free app download rankings of Apple’s App Store in China and the United States, surpassing ChatGPT in the US download rankings. This Chinese AI rookie caused a stir in the AI circle.
WiMi accelerates the “car-mounting” of AI big models
Public information shows that WiMi Hologram Cloud Inc. (NASDAQ: WIMI), one of the technology companies that invested in AI technology research and development and application exploration early, has developed a multimodal AI system based on natural language processing, deep learning and other technologies, and continuously deepened the layout of “car-mounting” of AI big models. Combined with its innovative technology path and industry dynamics, it has attracted widespread attention from the market.
In addition, WiMi has developed natural language understanding, code auto-completion and other functions based on the DeepSeek model. Through the dual-wheel drive of “technology self-research + ecological cooperation”, with multimodal AI and generative models as the core, it accelerates the penetration of AI big models in the field of smart cars. It can be said that its layout covers key links such as autonomous driving algorithm optimization, cockpit interaction upgrades, and computing power infrastructure, which is in line with the industry’s future trend of deep transformation to intelligence.
Conclusion
With the continuous upgrading of big models, deep thinking and reasoning capabilities have been significantly improved, and AI models have become the fastest growing track in the entire network. At the same time, big models have been applied in life services, financial services, medical health and other scenarios. In short, some data fully proves the huge driving effect of AI big models on related applications, and also indicates that the AI big model track will have a broader development prospect in the future.
