DeepSeek has been popular for 28 days. What are the domestic large-scale model peers doing?

Source: Zhidongxi

Zhidongxi reported on February 16 that the explosive popularity of Hangzhou-based deep search company DeepSeek has forced a number of leading large-model peers to urgently adjust their strategic layouts through open source. Baidu, Alibaba, Tencent, and NetEase have connected many of their national-level products to DeepSeek. Douyin Group has vigorously promoted Volcano Engine and Feishu. Three of the "Six Little Tigers of Big Models" have connected their products to DeepSeek.

In just these three days, Tencent, Baidu, and NetEase have added new actions one after another:

• On Saturday, WeChat, a super app with nearly 1.4 billion monthly active users and a high-quality private content pool, began grayscale testing of AI search access to DeepSeek's deep thinking capabilities;

• On Sunday, QQ Browser announced that it has connected to the full version of DeepSeek-R1. Baidu Search and Baidu Wenxin Intelligent Platform, which have 700 million monthly active users, also announced that they will fully connect to DeepSeek.

Today, Tencent Docs announced that it has officially connected to the full version of DeepSeek-R1. Tencent Cloud announced that Tencent Cloud AI Code Assistant, Tencent Element, Tencent LeXiang, and Tencent Maps have connected to DeepSeek. NetEase Fuxi announced that DepSeek has officially entered the mobile game "Ni Shui Han".

Three of the “Big Model Six Little Tigers” unicorns have actively embraced DeepSeek: Zero One Everything and MiniMax have respectively connected DeepSeek R1 to their overseas versions of the product, and Step Star has taken the lead in connecting its domestic flagship AI product Yuewen to DeepSeek R1.

Currently, Tencent's ima.copilot, QQ Music, Tencent Yuanbao, WeChat, QQ Browser, Tencent Documents, Tencent Cloud AI Code Assistant, Tencent Yuanqi, Tencent Lexiang, Tencent Maps, Baidu's Wenxiaoyan, Apollo Autonomous Driving, Baidu Maps, Baidu Search, Wenxin Intelligent Platform, Baidu Intelligent Cloud's Keyue, Xiling, Yijian, Zhenzhi, Alibaba's Tongyi Lingma, Dingding, Ant's Alipay Treasure Box, Byte's Feishu, 360's Nano AI Search, NetEase's Youdao Dictionary, "Ni Shui Han" Mobile Game, these products are all connected to DeepSeek.

Mobile phone brands that do not constitute a direct competitive relationship with DeepSeek have no worries about providing DeepSeek model options in their own AI assistants. Eight major mobile phone brands Huawei, Honor, OPPO, vivo, Meizu, Nubia, Red Magic, Lenovo Motorola have all announced that their AI assistants will be connected to DeepSeek-R1. Xiaomi was also exposed on February 14 that the "turn on deep thinking" function of Xiaomi Pengpai OS2 Super Xiaoai can use DeepSeek-R1, and then the function was temporarily offline.

Interestingly, although DeepSeek-R1 has stirred up a storm in the industry, the top featured articles on DeepSeek's official public account are still the DeepSeek-R1-Lite preview version on November 20 and the DeepSeek-V3 released on December 26.

The "trigger" that completely caused a huge shock in the global large model industry was the big move officially released on January 20 - DeepSeek-R1 performance benchmark OpenAI o1 official version + open source model weights + open thinking chain output + web version and App free to use + API input for only 1 yuan per million tokens + all paper training technology is open.

During these days when the spotlight was on, DeepSeek has kept a low profile. In addition to publishing a Spring Festival AI greeting article on January 27 and open-sourcing the multimodal unified understanding generation model Janus-Pro on January 28, it only updated an article on February 6 to combat fake and counterfeit accounts, "Instructions on DeepSeek's official information release and service channels."

From February 7th to 11th, DeepSeek's recruitment page released multiple social recruitment positions, including deep learning researchers [urgent], core system R&D engineers [urgent], full-stack development engineers [urgent], deep learning R&D engineers [urgent], data experts, senior operation and maintenance development engineers, senior technical experts (50-100K/month, etc.), data center technical supervisors, senior system operation and maintenance engineers, recruitment managers (20-60K/month), senior data acquisition engineers (25-50K/month), AGI legal affairs, etc. ‍

In terms of personnel changes, the corporate credit inquiry platform Qichacha shows that on February 15, DeepSeek's affiliated company DeepSeek underwent industrial and commercial changes. Pei Tian was changed from executive director and general manager to manager and director, and Wang Xianzu was added as the financial director. At the same time, Internet information services were added to the business scope.

Faced with the strong rise of DeepSeek, other large model peers may have spent an extremely memorable Spring Festival and post-holiday work.

Next, let's review what the domestic mainstream basic large language model suppliers have done in the past four weeks since DeepSeek-R1 was launched on January 20 (the names of the self-developed large models of these companies are in brackets).

01.7 large model unicorns:4 are connected to DeepSeek and can be experienced ‍‍‍‍‍‍‍‍‍

Before the Spring Festival, MiniMax, Step Star, Dark Side of the Moon, and Baichuan Intelligence all released new models, and the latter three all released inference models.

After the holiday, the "Big Model Six Little Tigers" have different movements. Step Star was the first to connect DeepSeek to domestic products, while Zero One Everything and MiniMax launched new DeepSeek in overseas products. Among them, Step Star and MiniMax are both focusing on multimodality, and Zero One Everything will shift its focus to lightweight models and AI-First applications this year, so it does not constitute a strong competitive relationship with DeepSeek.

Among the other three, Zhipu is busy releasing To B intelligent cooperation for AI PCs and smartphones, and Baichuan Intelligence focuses on medical models and products. The Dark Side of the Moon, which invested heavily in Kimi last year, only published its first WeChat article in the Year of the Snake today. According to a report by Huxiu on February 15, the Dark Side of the Moon has identified "continuously obtaining SOTA results" as the most important work goal at the moment.

In addition, APUS Kirin Hesheng, a mobile Internet unicorn with self-developed basic large models, has also announced access to DeepSeek.

1. Step Star

On January 20, the extremely fast-response, cost-effective lightweight inference modelStep-2 mini, as well as the Step-2 Literary Master Edition designed for creation, were released. The Step-2 Literary Master Edition has been launched on the Yuewen App, and the web version was fully launched in the week of the official announcement.

On January 21, it was announced that the multimodal understanding model Step-1o Vision was launched, and the voice model Step-1o Audio was upgraded. Both models are fully available on the Yuewen web version and App.

On January 22, the video generation model Step-Video V2 was released, supporting trial applications on the Yuewen web version.

On January 24, it was announced that the "Yuewen AI Creative Board" was launched, which was launched on the latest version of Yuewen, supporting 0 code foundation and 3-step application development.

Before February 16, the Step Star Yuewen web version will be connected to DeepSeek, which can be experienced for free and supports deep thinking, search, and file upload.

Portal: yuewen.cn/chats/

2. MiniMax Xiyu Technology (ABAB)

On January 20, it announced the newly upgraded T2A-01series voice model, which was launched on Conch Voice products globally, with a total of 300+ preset tones in different languages, supporting custom tones for transparency, strength, and superimposed echo, broadcast, distortion, electronic music and other special effects.

On February 16, it was announced that the overseas web version of MiniMax Chat has been connected to DeepSeek R1, with a limited time free experience, and does not support online search (selecting MiniMax-Text-01 supports online search).

Portal: chat.minimax.io

3. Yi

On January 23, the company released its 2024 year-end summary, announcing that it will focus on lightweight models, accelerate AI-First application exploration, actively embrace applications, and penetrate the industry.

On February 7, the company released an article in People's Daily interviewing Dr. Kai-Fu Lee, CEO of Yi, emphasizing that 2025 will be the first year of the explosion of AI-First applications.

On February 12, the company announced that the overseas version of its product PopAi will be connected to the DeepSeek R1 model, providing deep thinking and networking functions, and both the web version and the App are now online.

Portal: popai.pro

On February 14, it was announced that the industrial large model base jointly established by Zero One Everything and Suzhou High-tech Zone was officially awarded. As a large model base focusing on vertical industries, the base will focus on creating industry large model solutions in multiple fields such as manufacturing, finance, medical care, and government affairs.

4. APUS Kirin Hesheng (Tianyan)

On February 15, it was announced that the DeepSeek APUS enhanced large model was launched on the official website. This version is based on the full-blooded DeepSeek R1 large model, and is comprehensively optimized and improved from the three aspects of "usable, usable, and easy to use".

Portal: chat.apusai.com

5. Zhipu (GLM)

On January 21, it was announced that Qingying 2.0 is now fully launched on Zhipu Qingyan, bringing stronger video generation capabilities. Ordinary users can try it for free on Zhipu Qingyan, and members can enjoy the fast track (unlimited mode without deducting points).

On January 23, it was announced that the first computer agent GLM-PC for the public, which can be used as soon as the press enter, is open for experience. The latest "deep thinking" mode has been launched, and functions dedicated to logical reasoning and code generation have been added, and support for Windows systems has also been provided. Zhipu is conducting in-depth cooperation discussions with PC manufacturers such as Lenovo and Asus to promote the innovation and development of AI PCs.

On February 11, it announced that it would cooperate with Samsung based on Agentic GLM to bring the Agent experience to Samsung's latest mobile phone Galaxy S25 series.

On February 11, an old news from January 23 was released: Zhipu CEO Zhang Peng attended the Bank of China's "Action Plan to Support the Development of the Artificial Intelligence Industry Chain" release event and delivered a speech; Zhipu is the intelligent scene construction partner of the Bank of China Head Office and works closely with the Bank of China Beijing Branch.

6. Baichuan Intelligence

On January 24, it released the full-scene deep reasoning modelBaichuan-M1-preview and the medical enhancement open source model Baichuan-M1-14B.

On January 26, it was announced that the open sourceomnimodal understanding generation unified model Baichuan-Omni-1.5 was launched, and two evaluation sets OpenMM-Medical and OpenAudioBench were open sourced.

On February 14, it was announced that the first AI pediatrician was officially "on duty": Beijing Children's Hospital affiliated to Capital Medical University, National Children's Medical Center, conducted the first "AI pediatrician + multidisciplinary experts" dual-doctor parallel multidisciplinary consultation in China on February 13, introducing the "AI pediatrician" jointly developed by Beijing Children's Hospital, Baichuan Intelligence, and Xiaoerfang Health Technology based on the Baichuan M1 large model.

7. Dark Side of the Moon (Kimi)

On January 20, the company released a new SOTA model, the k1.5 multimodal thinking model, claiming that k1.5 "should be the first company outside of OpenAI to achieve the multimodal reasoning performance of the official version of o1 in the world", and released the model training technical report for the first time, announcing that the preview version will be gradually released on the web version and the latest version of the App.

On January 21, it was announced that Kimi Smart Assistant will be launched on the "2024 Cognitive Talent Report".

On January 26, it shared some annual stories of users using Kimi and announced Kimi's interactive report card (answers to 4 billion questions in 2024).

On January 31, it was announced that the latest iOS and Android apps and web versions of Kimi Smart Assistant will be launched on k1.5.

On February 17, the official account published an article to share the complete thinking process behind the k1.5 model answered by Flood Sung, a researcher of the Dark Side of the Moon, on Zhihu.  

02.15 technology giants with self-developed large language models have been connected to DeepSeek R1

Technology giants are all tough guys. Their advantage lies in their many independent businesses, so they don’t have to care too much about the feelings of their own large model teams. The key is to see who can seize the opportunity to play their own advantages and win more users and revenue.

The core business connected to DeepSeek is divided into two major parts, one focusing on providing computing power + model development services, and the other focusing on applications.

In terms of models, Ali Tongyi and Byte Doubao have released a number of new models.

At the all-staff meeting held on February 13, ByteDance CEO Liang Ruobo reflected that compared with startups, ByteDance's sensitivity to new opportunities in large models is obviously insufficient. He proposed that ByteDance's key goals in the field of AI this year include pursuing the upper limit of "intelligence", exploring new interactions and strengthening scale effects. He believes that compared with pursuing a specific product, taking intelligence itself as the most important goal can inspire more attempts and not ignore key technical nodes.

In terms of cloud services, Huawei Cloud fired the first shot of DeepSeek on the cloud, and Tencent Cloud, Baidu Smart Cloud, Alibaba Cloud, Volcano Engine, Inspur Cloud and others followed suit. Ant also announced that Alipay Treasure Box supports building intelligent entities with DeepSeek.

In terms of applications, Tencent and 360 have respectively connected their own AI large model products Tencent Yuanbao and Nano AI to the full-blooded version of DeepSeek-R1. Baidu Wen Xiaoyan is more subtle and does not directly provide the option of replacing the Wenxin large model. It only uses the R1 model to optimize the photo-taking problem-solving function.

QQ Music, QQ Browser, Tencent Maps, Baidu Maps, NetEase Youdao Dictionary and other national applications have been connected to DeepSeek. WeChat, which has the top 1 monthly active users in the country, has just started the DeepSeek grayscale test. I wonder if applications such as NetEase Cloud Music, Baidu Browser, AutoNavi Maps, and QQ will follow soon.

Some mobile phone giants with self-developed large language models, including Honor, OPPO, vivo, ZTE (Nubia and Red Magic), Lenovo Motorola, have connected their AI assistants to DeepSeek.

Because the update news of major manufacturers is too intensive, here we only sort out their related actions in large models, and do not involve the tutorials and partner cases they share.

1. Huawei (Pangu)

Huawei Cloud:

• On February 1, it was announced that it would jointly launch DeepSeek R1&V3 inference services based on Ascend Cloud with Silicon Mobility.

• On February 12, it was announced that the full-blooded version of DeepSeek V3/R1 would be launched, and the DeepSeek local deployment solution based on hybrid cloud would be launched.

Huawei Computing:

• On February 5, the DeepSeek model was announced to be launched on the Ascend community.

• On February 10, the DeepSeek V3 Ascend MindSpore version was announced to be launched on the open source community.

• On February 13, the full-blooded version of DeepSeek V3/R1 and the recommended configuration of the distillation model Ascend all-in-one machine were released.

Huawei Data Storage:

• On February 6, it was announced that Huawei ModelEngine supports local deployment of the entire DeepSeek series.

• On February 8, it was announced that Huawei OceanStor A800 supports DeepSeek hard disk cache technology.

• On February 16, the recommended configuration of Huawei DeepSeek training/promotion hyper-converged all-in-one machine was released.

Hongmeng version of Xiaoyi Assistant:

• On February 5, it was announced that the native Hongmeng version of Xiaoyi App was connected to the DeepSeek-R1 beta version.

• On February 10, it was announced that the DeepSeek-R1 on the native Hongmeng version of Xiaoyi App was upgraded to the official version, with a new online search function.

2. Tencent (Hunyuan)

Tencent Hunyuan:

• On January 21, it was announced that 3D Generation Large Model 2.0 was open source, and the industry's first one-stop 3D content AI creation platform Hunyuan 3D AI creation engine was launched simultaneously. You can apply for the experience on the Hunyuan 3D official website.

Tencent Cloud:

• On February 2, it was announced that the DeepSeek-R1 large model can be deployed to Tencent Cloud HAI with one click, and developers can access and call it in just 3 minutes.

• On February 4, it was announced that the Tencent Cloud TI platform launched the "Developer Gift Pack": one-click deployment of the entire DeepSeek model series, and some models are free to experience.

• On February 8, Tencent Cloud launched the API interface of DeepSeek-R1 and V3 original models, connected them to the large model knowledge engine, and supported online search, becoming the first cloud vendor in China to achieve this capability.

• On February 11, it was announced that developers can connect the full version of DeepSeek to WeChat mini-programs with at least 3 lines of code, and new cloud development users can also enjoy the first month of free packages and 1 million tokens.

• On February 17, it was announced that Tencent Cloud AI Code Assistant, Tencent Yuanqi, Tencent Lexiang, and Tencent Maps will be connected to DeepSeek. Tencent Cloud AI Code Assistant and Tencent Element are all free and unlimited; Tencent LeXiang AI Assistant will open all services based on the DeepSeek model for free; Tencent Maps is the first to integrate DeepSeek-R1 capabilities with the "location extraction" function, supporting users to generate and query travel guides with one click.

ima.copilot:

• On February 11, it was announced that Tencent's AI smart workstation ima was officially connected to the DeepSeek-R1 model, which can be experienced on Windows, Mac and ima knowledge base applet.

QQ Music:

• On February 11, it was announced that it would become the first music platform in China to integrate DeepSeek.

Tencent Yuanbao:

• On February 13, it was announced that DeepSeek R1 was connected to the Internet and fully launched, and both the web version and the App can be experienced for free.

WeChat:

• On February 15, WeChat Soso began grayscale testing of AI search access to DeepSeek.

QQ Browser:

• On February 16, it was announced that it had officially accessed the full-blooded version of the DeepSeek-R1 model, supporting deep thinking, online search, multi-round dialogue, and historical record backtracking.

3. Baidu (Wenxin)

Wenxin Yiyan:

• On February 13, it was announced that Wenxin Yiyan will be completely free from 0:00 on April 1, and all PC and App users can experience the latest models of the Wenxin series; starting from now, Wenxin Yiyan will launch the deep search function; this function will also be free to use from April 1.

• On February 14, it was announced that the Wenxin large model 4.5 series will be launched in the next few months, and will be officially open sourced on June 30.

Wen Xiaoyan:

• On February 11, the version was updated to access the DeepSeek-R1 model and optimize the photo-solving function.

Baidu Smart Cloud:

• On February 3, it was announced that the DeepSeek-R1 and DeepSeek-V3 models were available on the Baidu Smart Cloud Qianfan platform, and an ultra-low price plan was launched, and limited free services can also be enjoyed.

• On February 8, it was announced that Baidu Smart Cloud's four large model application products, Keyue, Xiling, Yijian, and Zhenzhi, officially launched a new version of the DeepSeek model.

• On February 8, it was announced that Baidu Baige AI heterogeneous computing platform has supported the rapid deployment of the full-blooded version of DeepSeek+ online search.

• On February 13, it was announced that Baidu Smart Cloud Qianfan AppBuilder has completed the full access to the DeepSeek model, not only with online search and other component capabilities, but also RAG, Agent, workflow and other functions have achieved full-link support for the DeepSeek model.

Wenxin Intelligent Platform:

• On February 16, it was announced that it would fully integrate DeepSeek.

Apollo Intelligent Driving:

• On February 13, it was announced that the Apollo smart cabin model and travel intelligent body had completed DeepSeek integration.

Baidu Map:

• On February 15, it was announced that the deep integration of DeepSeek-R1 has been completed.

Baidu Search:

• On February 16, it was announced that the latest deep search functions of DeepSeek and Wenxin Model will be fully integrated.

4. Alibaba (Tongyi)

Alibaba Cloud Tongyi Qianwen:

• On January 27, the latest paper of the Alibaba Cloud Tongyi Qianwen Qwen team was published, revealing the key details of Qwen MoE.

• On January 27, the open source Qwen2.5-1M model and its corresponding reasoning framework support were launched, extending the context of the open source Qwen model to 1M length for the first time.

• On January 28, the open-source visual understanding model Qwen2.5-VL was released, which can be experienced in Qwen Chat.

• On January 29, the ultra-large-scale MoE model Qwen2.5-Max was released, which can be experienced for free in Qwen Chat.

• On February 5, it was announced that Tongyi Qianwen Qwen2.5-Max ranked first in mathematics and programming in the latest large model blind test list released by the third-party benchmarking platform Chatbot Arena.

• On February 7, it was announced that Qwen2.5-Max ranked ninth in the world in the latest ranking of LiveBench, the authoritative model evaluation list, and is the best performing Chinese model among non-inference models.

• On February 10, it was announced that Tongyi Qianwen Qwen derivative models took the top ten in the world in HuggingFace's latest open source large model list.

• On February 13, Alibaba co-founder and chairman of the board of directors, Joseph Tsai, confirmed that Apple chose to cooperate with Alibaba to develop AI features for the Chinese version of the iPhone.

Alibaba Cloud:

• On January 29, it was announced that the Tongyi Qianwen flagship model Qwen2.5-Max can directly call API services through the Alibaba Cloud Bailian platform.

• On February 3, it was announced that Alibaba Cloud PAI Model Gallery supports one-click deployment of DeepSeek-V3 and DeepSeek-R1 on the cloud.

• On February 9, it was announced that the DeepSeek full-size model was launched on Alibaba Cloud Bailian and one million free tokens were given away.

Tongyi Lingma:

• On February 12, the model selection function was announced to support the full-blooded version 671B model of DeepSeek-V3 and DeepSeek-R1 based on Bailian.

DingTalk:

• On February 6, it was announced that DingTalk AI Assistant has fully accessed the DeepSeek series of models, supporting one-click creation and release of AI assistants based on DeepSeek models.

• On February 6, it was announced that DeepSeek was adapted to the DingTalk AI all-in-one machine.

• On February 7, it was announced that DingTalk YiDa was connected to DeepSeek.

5. Byte (Skylark)

Doubao:

• On January 20, the Doubao real-time voice model was launched and fully available in the Doubao App.

• On January 22, Doubao Big Model 1.5 Pro was officially released, and a technical report was released.

• On February 10, the general video generation experimental model VideoWorld was announced to be open source.

• On February 12, the sparse model architecture UltraMem was released, and the inference cost can be reduced by up to 83% compared with MoE.

Volcano Engine:

• On January 20, it was announced that Doubao Big Model 1.5 has been fully launched on Volcano Ark.

• On February 4, it was announced that the full-size DeepSeek model has landed on Volcano Engine, which can be deployed in the Volcano Engine machine learning platform veMLP, and the model can also be called in Volcano Ark. It also provides the highest flow limit in the entire network and a special two-week limited 50% discount.

Feishu:

• On February 10, it was announced that through tools such as Feishu multi-dimensional tables and Feishu intelligent partner creation platform, enterprises can have one-stop access to mainstream large models such as Doubao, DeepSeek, and Zhipu GLM.

6. Inspur (Hairuo)

Inspur Cloud:

• On February 7, the Hairuo all-in-one machine with pre-installed DeepSeek-R1 and DeepSeek-V3 models was released, which was the first to realize the domestic GPU single-machine inference service of the 671B DeepSeek large model.

7. Ant (AntGLM)

Ant Technology:

• On February 10, the Alipay Treasure Box intelligent agent development platform was connected to the full-blooded version of DeepSeek R1, which allows for free and unlimited construction of intelligent agents.

8. NetEase (Yuyan)

NetEase Cloud Commerce:

• On February 8, it was announced that it would fully integrate with DeepSeek to accelerate the inclusive application of large AI models in service and marketing scenarios.

Youdao Dictionary:

• On February 10, it was announced that Youdao Xiao P would fully integrate with DeepSeek, greatly improving its writing ability.

• On February 14, it was announced that Youdao Dictionary has fully connected to DeepSeek R1, a full-blooded version with zero lag, and no server busy issues, which will comprehensively improve the translation and learning experience.

NetEase Fuxi:

• On February 13, it was announced that the full-blooded version of DeepSeek was officially launched, helping NetEase Shuzhi's NetEase Shufan Youshu ChatBI, NetEase Yidun, NetEase CodeWave, NetEase Cloud Letter and NetEase Cloud Commerce and other businesses to fully connect to DeepSeek, and at the same time open NetEase Fuxi's self-developed MoA expert collaborative intelligent body.

• On February 17, it was announced that with the help of NetEase Fuxi [Youling·Yishengzhuxiang] platform, the intelligent NPC "Shen Qiusuo" driven by the official version of DeepSeek will officially enter the arena of "Ni Shui Han" mobile game. This is the first time that DeepSeek has entered the game, the first time it has a virtual image, and the first time it has become a virtual intelligent life form.
  

9. 360 (360 Smart Brain)

360:

• On February 2, it announced the launch of the "DeepSeek version" security model, becoming the first digital security company in China to access DeepSeek.

Nano AI Search:

• On February 11, it was announced that the "Nano AI" PC version client was launched, and the high-speed dedicated line and networked full-blood version of the DeepSeek-R1 model can be selected.

10. Lenovo (Tianxi)

Lenovo Xiaotian:

• On February 14, it was announced that Lenovo’s motoAI mobile phone was connected to the full-blooded version of DeepSeek-R1, supporting online search, and the user can ask questions through direct voice dialogue with Lenovo Xiaotian.

Lenovo Intelligent Cloud:

• On February 14, Lenovo Intelligent Cloud (xCloud) solution announced that DeepSeek R1 and its distillation series large models will be fully integrated into the product matrix, covering core links such as model services, model development, intelligent body construction, and intelligent operation and maintenance. Lenovo Zhixiaoxing has integrated the DeepSeek R1 large model, further enhancing the full-stack intelligent operation and maintenance capabilities.

11. Honor (Magic)

Honor YOYO Assistant:

• On February 8, it was announced that it would be officially connected to DeepSeek. You can upgrade YOYO Assistant to the new version and experience the DeepSeek-R1 trial version in the intelligent body section.

• On February 8, it was announced that the DeepSeek network version would be launched.

12. OPPO (AndesGPT)

ColorOS Xiaobu Assistant:

• On February 8, it was announced that ColorOS will fully integrate with DeepSeek. Xiaobu Assistant will be the first system intelligent assistant deeply integrated with DeepSeek, supporting voice to wake up DeepSeek answers in one step. The new version will gradually cover devices equipped with ColorOS 15, and will be available out of the box on the upcoming folding screen mobile phone OPPO Find N5.

13. vivo (Blue Heart)

OriginOS Blue Heart Xiao V:

• On February 12, it was announced that vivo will deeply integrate the full-blooded version of DeepSeek.

14. ZTE (Nebula)

Red Magic:

• On February 10, it was announced that the Red Magic phone will soon be connected to the DeepSeek large model.

Nubia:

• On February 13, it was announced that ZTE Nubia Nebula AI was the first to achieve full-size embedded DeepSeek-R1 and officially launched it on the Nubia Z70 Ultra model. The phone is not only equipped with a complete 671 billion parameter DeepSeek-R1 model, but also achieves system-level interactive integration.

15. Xiaomi (MiLM)

Surge OS Super Xiao Ai:

• On February 14, it was revealed that in Xiaomi Surge OS2, DeepSeek-R1 can be activated by saying "Open Deep Thinking" to Super Xiao Ai. It is said that this function is in the debugging stage and is temporarily offline, and will be back online later.

03.Other large-model listed companies:Selling computing power, releasing models, and receiving products

Among other listed companies with self-developed basic large models, iFlytek and SenseTime mainly provide DeepSeek model services through the MaaS model, Kunlun Wanwei, Mobvoi, and Zhihu have connected DeepSeek to their own products, and Yuntianlifei and CloudWalk Technology have both announced that their training and promotion machines have been successfully adapted to DeepSeek.

1. iFlytek (Xinghuo)

On February 10, it was announced that the entire series of DeepSeek large models were officially launched on the iFlytek open platform, supporting public cloud API calls and one-click deployment of exclusive models. From now until March 10, the DeepSeek reasoning API will be open for free.

2. SenseTime (Ririxin)

On January 20, it was announced that SenseTime’s Ririxin fusion large model interactive version (SenseNova-5o) officially provides real-time audio and video dialogue services, which are free for a limited time.

On January 22, it was announced that SenseTime’s “Miaohua Fun Shoot” App was launched, and 500 permanent points were obtained upon first login.

On January 26, it was announced that SenseTime’s Ririxin fusion large model interactive version (SenseNova-5o) ranked first in China with a total score of 68.59 in the first test list of the real-time audio and video Chinese evaluation benchmark SuperCLUE-Live.

On February 7, it was announced that the consortium with Guangdong Mobile won the bid for the “AI Middle Platform Large Model Capacity Building Project” of Guangzhou Yuexiu Enterprise Group.

On February 8, it was announced that SenseTime’s large device was launched on the DeepSeek series of models, and corporate customers and developers can enjoy the right to use 10 million tokens for free within 3 months.

3. Kunlun Wanwei (Tiangong)

On February 8, it was announced that the PC version of Tiangong AI was updated and the "DeepSeek R1 + Online Search" function was launched.

Portal: tiangong.cn

On February 14, the Matrix-Zero world model was launched, and Kunlun Wanwei became the first company in China to launch 3D scene generation and interactive video generation models to explore spatial intelligence.

On February 15, it was announced that Tiangong AI ranked second in the overall list, second in complete response rate, and second in accuracy in the "DeepSeek-R1 Web Stability Evaluation Report" released by the third-party AI evaluation agency SuperCLUE.

4. Mobvoi (Sequence Monkey)

On February 7, it was announced that it had completed deep adaptation with DeepSeek-R1 and integrated its advanced capabilities into the company's full range of AIGC products.

5. Zhihu (Zhihaitu AI)

On February 11, it was announced that Zhihu's AI search product Zhihu Direct Answer was officially connected to the full version of DeepSeek R1, and was integrated and launched on both the Zhihu Direct Web page and the Zhihu App.

Portal: zhida.zhihu.com/pro

6. Yuntian Lifei (Tianshu)

On February 5, Yuntian Lifei’s chip team announced that it had completed the adaptation of the DeepEdge10 “computing power building block” chip platform to the DeepSeek-R1 1.5B, 7B, and 8B distillation models. The R1 32B distillation model and the V3/R1 671B MoE large model are also being adapted in an orderly manner. After the adaptation is completed, DeepEdge10 will fully support the full range of DeepSeek models in the end, edge, and cloud.

On February 14, it was announced that Yuntian Lifei training and push-pushing all-in-one machine was successfully adapted to DeepSeek and launched in Longgang, Nanshan and other places in Shenzhen.

7. Cloudwalk Technology (Congrong)

On February 10, it was announced that it would jointly launch the first DeepSeek end-side office equipment, Zhongxun AI Mouse, with Zhongshu Xinke.

On February 13, it was announced that Cloudwalk Technology's Congrong large-scale model training and push-pushing all-in-one machine was successfully adapted to DeepSeek, which can achieve out-of-the-box private deployment.

On February 13, it was announced that Xunzhi AI·Intelligent Body All-in-One Machine is adapted to domestic computing power, pre-installed with mainstream large models such as DeepSeek, and the AI ​​intelligent body platform is ready to use out of the box.

04.Conclusion: What a great era of open source for big models

Some of DeepSeek's big language model peers followed the trend, some stayed put, some joined in large numbers, and some were in a state of panic.

Step Star, Tencent, and 360 have all added DeepSeek R1 deep thinking + online search options to their own AI big model products. Alibaba and Doubao have updated their own models the most frequently. Tencent and Baidu have multiple products from different businesses connected to DeepSeek. Baidu also announced that it will open Wenxin Yiyan and open source Wenxin Big Model 4.5 for free.

Kimi and Doubao, which had the most investment last year, seem to regard achieving the best model performance as the focus of this year's senior management, and are ready to compete with DeepSeek in terms of models.

Over the weekend, WeChat and Baidu Search, two major portals with huge traffic, joined DeepSeek, shocking the entire market. Alibaba and ByteDance have been relatively restrained in their product access. The first to embrace DeepSeek were cloud services and collaborative office software. Alibaba's Taobao, Alipay, AutoNavi, and ByteDance's Douyin, these super apps with high monthly activity, have not announced full access to DeepSeek.

Next, how will the "Big Model Six Little Tigers" tell a good new story? Which big model supplier will be the next product to be launched on DeepSeek?