Elephants can dance, but they must learn to pivot swiftly.

Produced by | Xinpinlue Finance | Author | Wu Wenwu

Recently, DeepSeek has emerged as a disruptor in the global tech and AI landscape, akin to a giant catfish stirring up the waters. Against this backdrop, Baidu, China's pioneer in large model releases, appears to be playing a quieter role, occasionally becoming the butt of jokes.

Recently, media reports highlighted that in Beijing's Haidian District's Raycom InfoTech Center, Baidu Ventures occupies the 6th floor, while DeepSeek resides on the 5th. Soon after, the internet buzzed with jokes about Baidu missing out on investing in DeepSeek. Gao Xue, CEO of Baidu Ventures, responded lightheartedly, saying, "We're just upstairs, but QuantOpt doesn't need the money."

Addressing the impact of DeepSeek, Baidu executives recently stated that while DeepSeek may pose a short-term challenge, in the long run, the benefits outweigh the drawbacks.

On February 13, Baidu officially announced two major developments: Firstly, Wenyan Yixin will become entirely free starting April 1, granting all PC and app users access to the latest Wenxin series model. Secondly, Wenyan Yixin will immediately launch a deep search function, also free of charge from April 1.

On the same day, media outlets widely reported on Baidu's announcement of Wenyan Yixin's free access, the upcoming deep search function, and Robin Li's views at the 2025 World Governments Summit held in Dubai.

From an external perspective, while it might seem Baidu proactively made these announcements, in the context of DeepSeek's growing popularity, it appears Baidu is also feeling the pressure. It's evident that Baidu is being pushed to its limits.

01 A frontrunner in China's large model race, now facing stiff competition

In the grand wave of the global AI era, Baidu entered early and has been consistently running, emerging as a frontrunner in China's large model race.

Flashback to November 30, 2022, when OpenAI unveiled ChatGPT, stunning the world and igniting a global wave of large models. Many Chinese internet technology companies followed suit, with Baidu being the swiftest.

In March 2023, Baidu led the way by releasing the Wenxin large model, akin to the ChatGPT project, positioning itself as a frontrunner and leader in China's large model race.

Baidu's achievements in large model development are undeniable. Data shows that by early November 2024, the daily average call volume of Baidu's Wenxin large model had surpassed 1.5 billion times, a 30-fold increase from the 50 million disclosed a year earlier.

As domestic internet technology giants successively invest in large models, while Baidu maintains a clear leading edge, competition is intensifying, and Baidu is beginning to feel the heat from behind.

Large models and search businesses can seamlessly integrate, igniting an AI search war. Search is Baidu's core business, and naturally, Baidu is reluctant to lag behind, continuously advancing in AI search.

Last November, Tencent announced that its Sogou Input Method would introduce AI search and AI quick check functions, along with ima, an AI smart work platform integrating search, reading, and writing. Around the same time, 360 also launched a new product, Nano Search.

Additionally, Douyin is pushing forward in the search business, introducing independent apps such as Douyin Search and Doubao PC Search. Xiaohongshu, a popular community for young people, is also making strides in search and launched a new AI search product called Diandian last year.

Although Baidu had previously integrated AI intelligent response tools into its PC search engine, with competitors rapidly developing AI search businesses, Baidu seemed to lag significantly in this area, only launching its AI search product on January 2 this year, somewhat belatedly.

Several recent events have once again drawn market attention and focus on Baidu's prowess in large models and AI-related businesses.

Apple has been seeking domestic partners to integrate AI functionality into its products in China. Initially, Apple approached Baidu and later contacted Tencent, ByteDance, etc. Surprisingly, Apple ultimately chose to cooperate with Alibaba. Alibaba Chairman Joe Tsai officially confirmed the news of Alibaba's collaboration with Apple on February 13.

According to The Information, Apple initially selected Baidu as its primary partner. However, Baidu failed to meet Apple's standards in developing the Apple Intelligence model, leading to obstacles in their collaboration.

The reason behind Apple's decision not to choose Baidu is clear. Originally, Baidu was the most promising candidate for collaboration with Apple. If successful, it would have been a commercial testament to Baidu's comprehensive AI strength, but it did not materialize.

Apart from the impact of DeepSeek's popularity, Baidu is also affected by OpenAI's latest moves. Prior to Baidu's announcement that Wenyan Yixin would be free, OpenAI launched the o3-mini model, benchmarked against DeepSeek-R1, and made it available to free users.

Previously, Baidu's Wenxin large model 4.0 was charged, with a monthly membership fee of 59.9 yuan and an annual fee of 659 yuan. Without the rapid market changes in recent months, Baidu would probably not have opted to offer it for free.

As a frontrunner in China's large model race, while Baidu still holds a leading position, it is facing market changes much faster than anticipated. From being an initial frontrunner, Baidu is now being caught up with, pushed forward, and even forced to adapt, surpassed by the next wave. It is now facing stiff competition.

As Robin Li recently stated in his latest views on DeepSeek, innovation cannot be planned.

02 Elephants can dance, but they must pivot proactively

Essentially, there are two development models and paths for large models: open source and closed source. The debate between open source and closed source has persisted in the domestic large model community.

In the global AI large model race, OpenAI represents the open source approach, and Alibaba, which recently partnered with Apple, also follows the open source route.

Baidu previously adhered to the closed source path. In July 2024, at the World Artificial Intelligence Conference, Robin Li stated that open sourcing models do not achieve the effect of "many hands make light work," and commercial closed source models are the most competitive. Li also emphasized that closed source models can concentrate on computing power and talent, achieving profitability through commercial operations.

DeepSeek, which has gained sudden popularity in recent months, is open source and allows others to use R1 for data distillation to train their own models, seemingly impacting Robin Li's perception and clearly altering Baidu's stance.

This time, Baidu suddenly announced that Wenyan Yixin would be entirely free, citing the iterative upgrades and continuously decreasing costs of the Wenxin large model as the reason. While this rationale might seem "official," in the fiercely competitive large model environment, Baidu appears to have had no choice but to offer it for free to attract users.

Baidu is an established internet technology company and a massive enterprise with its own operating mode, decision-making process, and protocols, often leading to slower market responses. After all, it's challenging to make an elephant dance swiftly.

In contrast, startups like OpenAI and the suddenly popular DeepSeek can quickly update and iterate technology, respond rapidly to the market, and adjust market strategies.

In Xinpinlue Finance's view, Baidu genuinely needs to thoroughly transform its existing business mindset in developing AI large models, improve efficiency, and respond as swiftly as possible to the latest industry changes.

For instance, last year, giants like ByteDance, Alibaba, Tencent, and iFLYTEK ignited a price war for large models. At the end of 2024, Alibaba once again announced a significant price reduction for large models, with a reduction rate exceeding 80%, intensifying the price war.

Amidst the price war for large models, it's difficult for Baidu to remain aloof. It cannot avoid participating. Coupled with the latest market trends in large models in recent months, Baidu should have reduced prices significantly or even made them free much earlier.

The large model race is evolving at breakneck speed, and no one can predict the future. A fiercely competitive long-distance race for large models has only just begun. Despite being in the lead, Baidu must maintain a sense of crisis and urgency.

With Wenyan Yixin soon to be entirely free, Baidu can only temporarily forgo some income at this stage to attract users. The next focus for Baidu is to fully develop its B-end business in enterprise large models.

Meanwhile, in this AI search war, Baidu will undoubtedly face increasing competition and challenges from peers. It's evident that Baidu understands the stakes in AI search.

Baidu's proud intelligent driving business has also recently been impacted by BYD's full-line rollout of intelligent driving features. Baidu's intelligent driving business also needs to accelerate commercialization and develop its RoboTaxi service, Luobo Kuaipao.

To consolidate and maintain its leading position in the field of large models, the most critical point for Baidu is to present a new, game-changing large model.

Recently, CNBC quoted a source stating that Baidu plans to release the next-generation AI model, Ernie 5.0, in the second half of this year, with significant enhancements in the model's multimodal capabilities.

No one can predict what the large model race will look like in the second half of the year, but Baidu must accelerate its pace. Whether it can unveil a new version of the large model that amazes the market at that time can only be answered by time.

In summary, Baidu truly needs to step up its efforts!