<p>From last night to today, there have been three major news items in the AI circle, and the narrative of Chinese technology continues to strengthen.</p><h2>1) Alibaba Tongyi Open Source RL New Model</h2><p>Alibaba Tongyi did not go back on its word, saying it would open source a new RL model this week, and it was released last night. The most impressive part is that the 32B performance rivals that of the full-powered DeepSeek R1. In the AIME24 evaluation set testing mathematical abilities and the LiveCodeBench assessing coding abilities, the QwQ-32B performed comparably to DeepSeek-R1, far surpassing o1-mini and the same-sized R1 distilled model. It is now available for experience on the Tongyi APP and web version.</p><p><img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/920fb079-3ebf-4105-a6ec-826a298b7c5a.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/920fb079-3ebf-4105-a6ec-826a298b7c5a.png"/></p><p>Moreover, it seems that this RL training did not take too long. Friends from Alibaba reported that, unlike traditional reward models, this time feedback for mathematical problems was provided by verifying the correctness of generated answers.</p><p><img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/872f9358-5d9f-4f52-af4e-dc3a7e505b61.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/872f9358-5d9f-4f52-af4e-dc3a7e505b61.png"/></p><p>The threshold for local deployment of high-performance large models has further decreased; now a 4090 is sufficient, and a Mac mini with slightly more than 24G memory (Apple's architecture allows GPU memory sharing) can be deployed. With national subsidies, it can be done within four digits.</p><p>The implication is that the potential of RL has been amplified again, and the enhancement of intelligent hardware capabilities may come soon? The demand for applications will explode, which is also beneficial for the development of Agents.</p><h2>2) Mac Studio Can Fully Power DS-R1 Locally</h2><p>Apple's release of the Mac Studio at a price of over 70,000 can fully power DS-R1 locally, expected at 20 token/s; (The threshold for full-powered R1 localization has significantly decreased from over 700,000 to over 70,000) <img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/818a3690-af72-4020-bc3b-a539b09584cd.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/818a3690-af72-4020-bc3b-a539b09584cd.png"/></p><h2>3) Manus AI Flooding the Screens</h2><p>Manus AI, developed by the Monica team, is flooding the screens with its Agent (although it is obviously a promotional piece), but the market has gained more understanding and expectations regarding Agent capabilities, and expectations on the application side have risen higher. (Computing power, computing power, computing power);</p><p><img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/0dc796e8-50ba-432b-baea-91b8551c130d.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/0dc796e8-50ba-432b-baea-91b8551c130d.png"/></p><p>These three points will change some market narratives about AI, but it can be anticipated that the scaling law of large-scale reinforcement learning has already emerged, and Chinese technology companies represented by Alibaba will face continuous re-evaluationHard AI, original title: &#34;Three Things Last Night, Strengthening China's AI Technology Narrative?&#34;</p><p>Risk Warning and Disclaimer</p><p>The market has risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Investment based on this is at one's own risk</p>

DPSK

BABA

09988

00700

BIDU

09888

BYTED

<p>Last night, three major news items emerged from China's AI circle, further strengthening the narrative of Chinese technology. First, Alibaba Tongyi open-sourced a new RL model, with performance comparable to DeepSeek R1, and the threshold for local deployment has been lowered. Second, Apple's Mac Studio can now run DS-R1 locally, with a significant price drop. Finally, Manus AI's Agent has attracted market attention, raising expectations for Agent capabilities. These changes indicate that Chinese technology companies are set to undergo a continuous re-evaluation</p>

- Three significant developments in the AI sector highlight the strengthening narrative of Chinese technology.  
- Alibaba has released a new RL model, QwQ-32B, which matches the performance of DeepSeek-R1, while Mac Studio can now support full DS-R1 locally at a significantly reduced cost.  
- These advancements suggest an impending surge in demand for AI applications and a reevaluation of Chinese tech companies, particularly in the realm of reinforcement learning.

Last night, three events strengthened China's AI technology narrative?