China's artificial intelligence applications are continuously causing a stir in the global market. Following the success of the low-cost AI model DeepSeek, Manus - a multi-purpose AI assistant built by a Chinese development team has attracted the attention of the technology world when launching the world's first universal AI Agent.
With its powerful and profound application in many fields, this AI is considered the next DeepSeek. However, China still has two other AI models that are equally noteworthy, with support from Alibaba and Tencent.
The presence of these four AI products is helping the position of the billion-people country in the AI race to be higher than ever.
R1 (DeepSeek)
DeepSeek was the first Chinese name to shake up the AI industry when it released DeepSeek-V3. This model achieved performance on par with GPT-4o but used only a fraction of the computational resources.
Not long after, DeepSeek followed up with its DeepSeek-R1 reasoning model. According to TechCrunch , R1 outperformed OpenAI’s o1 on benchmarks like AIME, MATH-500, and SWE-bench Verified.
At the same time, one of DeepSeek's models cost just $5.6 million , compared to the hundreds of millions of dollars that leading US companies spend to train their models.
 |
DeepSeek-R1 gives the world a different perspective on China's AI. Photo: MIT Technology Review. |
Immediately after the news was announced, shares of US technology companies plummeted. On January 27, Nvidia shares fell 17%. According to Bloomberg , this was the largest market value drop in US stock history, with a loss of nearly $600 billion .
The company's success has prompted many international investment funds to reassess the value of Chinese technology companies, despite pressure from the US.
QwQ-32B (Alibaba)
Following DeepSeek, Alibaba - China's leading e-commerce and cloud computing company has publicly released its open-source AI QwQ-32B. According to the tech giant, the latest model has even surpassed the performance of R1.
Specifically, Alibaba said QwQ-32B was mined with only 20% of the parameters of DeepSeek-R1, but still delivered performance equal to or superior to its competitors in areas such as mathematics, programming, and general problem solving.
 |
The QwQ-32B model is claimed by Alibaba to outperform DeepSeek-R1. Photo: DataCamp. |
Under CEO Eddie Wu, the tech giant is refocusing on its core strengths, with cloud computing and artificial intelligence being key projects in its growth plan.
“Many hedge funds and long-term investors see AI as a turning point for Alibaba, some want to value the cloud business and the advantages of large language models.
The AI story is seen as a driver of the company's reassessment, but concerns remain about its AI monetization strategy,” JPMorgan analysts said.
Yuanbao (Tencent)
On March 5, Tencent's Yuanbao Chatbot surpassed DeepSeek to become the most downloaded iPhone app in China. The app integrates Tencent's own Hunyuan AI technology, along with DeepSeek's R1 reasoning model.
The event reflects the fierce competition in the AI field in the world's most populous country. Tech giants are racing to develop smart chatbots to attract users, Bloomberg commented.
According to SCMP , Yuanbao's big advantage is that the company behind Tencent, based in Shenzhen, operates China's largest social networking app WeChat, used by nearly 1.4 billion people.
 |
Tencent's Yuanbao chatbot surpasses DeepSeek to become the most downloaded iPhone app in China. Photo: App Store. |
Enabling Yuanbao downloads within WeChat has made the AI chatbot more accessible to the app's massive user base.
Currently, the details of this model are still a mystery as Tencent has not released clear information about its ability to directly compete in technical performance with R1.
Manus
Manus is introduced as an AI Agent product, developed by Butterfly Effect, based in Beijing and Wuhan (China).
In the commercials, it was introduced to analyze data by keywords. This AI also ranked home prices based on the needs and income of the owner. The tool was further tested in the task of researching Nvidia and Tesla stocks.
Typically, users interact with AI by making requests, and they respond. The process continues like that. Agents like Manus minimize the manipulation. After the initial command, it automatically searches and thinks about the next step to execute.
 |
Manus is billed as the world's first universal AI Agent. Photo: IDR. |
In other words, unlike chatbots that require users to enter additional input, Manus can complete complex tasks. Entire requests like filing, stock analysis, data collection, and even website building can be done with just a single command on this AI.
On its official website, Manus says its product scored significantly higher on the GAIA SOTA test than OpenAI’s product, a metric that shows that the AI is good at solving real-world problems.
However, researchers also pointed out many major problems with this AI. Manus is suspected to be just a copy of foreign tools.
Its core functionality is very similar to ComputerUse from Anthropic. The automated planning capabilities again call apis from large language models (like GPT-4). In the case of stock analysis, Manus just takes Yahoo Finance data, then charts it based on Python scripts.
Source: https://znews.vn/bo-tu-ai-trung-quoc-dang-lam-mua-lam-gio-the-gioi-cong-nghe-post1537727.html
Comment (0)