Subtle shifts are underway in the AI landscape.

By Han Yijia, Investor Network

The partnership between WeChat and DeepSeek has widened the horizons for domestic AI giants, propelling them into deeper explorations.

Recently, some WeChat users have noticed an 'AI Search' option in the WeChat Search bar. Upon selecting 'AI Search', users can choose between 'Quick Answer' and 'In-depth Thinking'. The latter mode, powered by the DeepSeek-R1 model, delivers more comprehensive answers after thorough consideration.

On February 16, Tencent confirmed that WeChat Search had officially commenced a gray-scale test to integrate DeepSeek. Users included in this test can see the 'AI Search' option at the top of the dialogue box. By clicking on it, they can utilize the full-featured version of the DeepSeek-R1 model free of charge.

The fusion of WeChat and DeepSeek has once again stirred up the AI industry's new horizon.

Tencent Embraces DeepSeek

Earlier this February, Tencent Cloud announced the launch of API interfaces for the DeepSeek-R1 and V3 original models. Simultaneously, the Knowledge Engine, a large-model knowledge application development platform under Tencent Cloud, integrated these two models and supports online search, becoming one of the first cloud vendors to support DeepSeek API + online search.

Half a month later, Tencent took a significant step forward, announcing the integration of DeepSeek into its nation-wide social software WeChat, officially embracing DeepSeek in depth.

Concerning this collaboration between the two major platforms, many users have raised concerns about whether DeepSeek will collect content from personal Moments and official accounts. Tencent has addressed this query.

As is well-known, the data content on the WeChat platform is not fully shared with other search engines and constitutes a part of Tencent's 'private domain' traffic. Therefore, integrating DeepSeek also signifies that Tencent will 'share' WeChat content with DeepSeek. Tencent clarified that the DeepSeek integrated into WeChat AI Search supports online search (without manual selection by users). Leveraging rich content from the WeChat ecosystem, including official accounts, and high-quality content from the entire network, it provides users with more comprehensive and high-quality answers.

Regarding whether this collaboration is fully implemented, Tencent stated that this capability is still in the gray-scale testing phase and will continue to be optimized based on user experience and feedback.

When asked why WeChat is integrating large models into its search scenarios, Tencent responded that large models can enhance search intelligence and accuracy, such as better understanding users' search intentions and analyzing complex query content. Aligning with user needs, Tencent has integrated large models, including Hunyuan and DeepSeek, into search scenarios to further enrich users' search experience.

Netizens are particularly concerned about whether AI Search will use their personal information, such as Moments and chats within WeChat. Tencent assured that AI Search only integrates public information from official accounts and other public sources on the internet and does not utilize users' personal information or related privacy details.

It is understood that multiple Tencent products are exploring the integration of DeepSeek, such as Tencent Cloud AI Code Assistant and Tencent Yuanbao, to offer users a richer experience and services.

On February 13, Tencent AI Assistant 'Tencent Yuanbao' announced support for both Tencent Hunyuan and DeepSeek models, utilizing the full-featured version of DeepSeek-R1671B to support online search and integrating Tencent ecosystem information sources like WeChat official accounts and video numbers.

On February 16, QQ Browser also announced the official integration of the full-featured version of the DeepSeek-R1 model, supporting in-depth thinking, online search, multi-round dialogue, and historical record traceability.

Tencent possesses its self-developed Hunyuan large model. By choosing to integrate DeepSeek-R1, it can swiftly implement AI search capabilities within the WeChat ecosystem. Analysts believe that Tencent's selection of the 'external technology + own scenarios' middle path not only avoids the high investment risks of self-developed large models but also swiftly compensates for shortcomings in AI capabilities. DeepSeek's natural language processing advantages can enhance the accuracy of WeChat search, where AI can provide personalized answers based on chat records and official account content. DeepSeek's open-source attribute is compatible with the WeChat ecosystem and can be expanded to scenarios such as intelligent customer service in the future.

DeepSeek's Growing Circle of Friends

The emergence of DeepSeek has profoundly influenced industry trends. Since February, platforms such as Baidu Intelligent Cloud, Huawei Cloud, Alibaba Cloud, Tencent Cloud, 360 Digital Security, and CloudAxis Technology have announced the launch of the DeepSeek large model, enabling users to call models like DeepSeek-R1 and DeepSeek-V3 on various platforms.

On February 1, Huawei Cloud's official WeChat account announced that the Silicon-based Flow and Huawei Cloud team jointly launched and went live with the DeepSeek R1/V3 inference service based on Huawei Cloud's Shengteng Cloud Service.

On February 3, Baidu Intelligent Cloud announced that the Baidu Intelligent Cloud Qianfan Platform had officially launched the DeepSeek-R1 and DeepSeek-V3 models, introducing an ultra-low-price plan. Users can also log in to Baidu Intelligent Cloud Qianfan ModelBuilder to enjoy free services for a limited time. Also on February 3, Alibaba Cloud announced that the Alibaba Cloud PAI Model Gallery supports one-click deployment of DeepSeek-V3 and DeepSeek-R1 on the cloud.

On February 4, JD Cloud announced that the Yanxi AI development and computing platform already supports one-click deployment of DeepSeek-V3, DeepSeek-R1, and all distilled small parameter models (DeepSeek-R1-Distill), supporting both public cloud online deployment and dedicated/hybrid private instance deployment modes for users to deploy and quickly call as needed. On the same day, ByteDance's Volcano Engine announced support for V3/R1 and other sizes of open-source DeepSeek models.

On February 5, Mobile Cloud announced the full launch of DeepSeek, achieving full version coverage, full size adaptation, and full functionality. China Mobile's 13 intelligent computing centers nationwide have fully launched these capabilities, allowing users to choose any intelligent computing resource pool for deployment, distillation, intelligent agent orchestration, and other operations. On the same day, China Unicom also announced that UnionCloud has achieved compatibility of domestic and mainstream computing power with multiple specifications of the DeepSeek-R1 model based on the Xingluo platform, catering to both privatization and public scenarios, providing comprehensive operational service guarantees.

On the evening of Tencent WeChat's integration of DeepSeek, Baidu promptly announced that Baidu Search and ERNIE Bot platform will also fully integrate DeepSeek and the latest deep search capabilities of the ERNIE large model. Search users can utilize the deep search capabilities of DeepSeek and the ERNIE large model for free, and developers on the ERNIE Bot platform will also be able to call the DeepSeek model at any time to create and optimize intelligent agents.

In just one month, including 'big factories' and platforms, they have swiftly joined DeepSeek's growing circle.

Not only in China but also overseas, major cloud and chip vendors have announced the launch of the DeepSeek-R1 model even earlier. On January 29, Microsoft announced that DeepSeek-R1 was available on Azure AI Foundry and GitHub, enabling developers to test the new model and build cloud-based applications and services. On January 31, NVIDIA announced the launch of the DeepSeek-R1 model on NVIDIA NIM. Amazon AWS and Microsoft Azure have also integrated the DeepSeek-R1 model.

The advent of open-source DeepSeek has undoubtedly impacted the development pace and landscape of the AI industry both domestically and internationally. The former 'Six Little Tigers' of AI (Lingyi Wanwu, Zhipu AI, Baichuan Intelligence, MiniMax, Dark Side of the Moon, and Step-by-Step Stars) have all taken new actions and responses.

Lingyi Wanwu and Alibaba Cloud have established a 'Joint Lab for Industrial Large Models', with some training and AI infrastructure teams joining the lab. Simultaneously, Lingyi Wanwu has clearly stated that it will no longer pursue training super large models but will focus on lightweight models with moderate parameters to achieve faster and cheaper commercialization.

MiniMax also released and open-sourced its new generation of 01 series models on January 15, 2025, supporting context input of up to 4 million tokens, far exceeding the 128,000 tokens of Kimi k1.5. Simultaneously, MiniMax is exploring monetization paths for overseas To C applications, with its star product Talkie becoming the fourth most downloaded AI application in the United States in the first half of 2024.

Dark Side of the Moon and Zhipu AI remain focused on achieving technological innovation and pursuing AGI, continuing to explore commercial monetization paths for To C applications. On February 16, according to media reports, Dark Side of the Moon has listed 'continuously achieving SOTA results' as its most important work goal at present. In 2025, Dark Side of the Moon will focus on enhancing multimodal and long-text reasoning capabilities in terms of model capabilities.

Zhipu AI CEO Zhang Peng expressed his commitment to 'All in AGI', believing that everything else is just a path and that optimizations are being made daily. Baichuan Intelligence is deeply involved in vertical industries, focusing on the application and implementation of large models in the medical field.

Step-by-Step Stars is still catching up in the development of basic models and recently released the Step-1o Vision multimodal understanding large model. It is understood that the Step-1o series is an end-to-end text, vision, and speech trimodal generation and understanding integrated model developed by Step-by-Step Stars, with commercial implementations primarily focused on the To B field.

The popularity of DeepSeek poses both technical and commercialization challenges to the 'Six Little Tigers' of AI. In the future, entrants into the AI industry will further segment their business paths. (Produced by Sifinance) ■

Source: Investor Network