Image source: Generated by Unbounded AI

In mid-March, Huawei reported that it had formed a medical and health corps, focusing on the clinical implementation of the medical model. This news caused a thousand waves within Baichuan Intelligent, which is the "Six Little Tigers of AI".

Huawei's entry has allowed Baichuan to welcome an absolute weight competitor at the AI+ medical field.

Medical care is the lifeblood of Baichuan's business today. "Intelligent Emergence" exclusively reported in early March that Baichuan Intelligent abolished the B-end group responsible for finance, education and other fields, with the reason for concentrating resources and focusing on the core business of medical care.

"The aftermath of DeepSeek is still there, and Huawei is coming again with a knife." An employee of Baichuan told "Smart Emergence".

Abolition of the B-end and focusing on medical care are Baichuan's decision to deal with DeepSeek's table-turning. Nowadays, how to avoid direct competition with Huawei on "B/G harvester" is a new proposition that Baichuan has been discussing quickly.

What is Baichuan Intelligent doing? Under Wang Xiaochuan's helm, where will Baichuan sail? This seemingly simple question has become increasingly confusing after Baichuan's business adjustments.

"Intelligent Emergence" interviewed several internal employees of Baichuan to try to restore a clear face of this AI company.

For a long time, there are four business groups that support Baichuan’s intelligent operation: production and research, general, B-end, and medical care. "Intelligent Emergence" learned that since its establishment in April 2023, Baichuan has experienced three strategic changes around these four groups: from focusing on model research and development and implementation of the B-end in the finance and education industry, to testing C-end products and multi-modal models; then transforming and focusing on medical care, B-end commercialization in parallel; finally abolition of the B-end and focusing on medical care - these adjustments are all to keep Baichuan at the card table.

The current Baichuan has the greatest confidence to stay on the Liu Xiaohu card table, which comes from cash flow.

At an internal meeting, Ru Liyun, co-founder of Baichuan Intelligent, divided Liu Xiaohu's cash flow level into 5 levels in his speech: those who can live for 6 months, 12 months, 24 months, 48 ​​months, and more than 48 months - Baichuan is in the "more than 48 months".

The year when C-terminal and multimodal disorders were not smooth

Until the first half of 2024, Baichuan Intelligent also had a clear development path: focusing on the research and development of underlying models, and at the same time taking ToB projects in the fields of finance and education as the pillar of commercialization.

As for C-end projects and AI medical products, Baichuan has tried the waters on a small scale internally. An employee of Baichuan recalled that Wang Xiaochuan also attaches importance to the C-end, but his goal is far more than a question-and-answer product, but slowly polishing out a "super application".

However, the rapid changes in the industry require companies to submit C-end answer sheet as soon as possible - early 2024, monthKimi's appearance on the dark side gave the market a dream of AI To C. At that time, Q&A products that support long text understanding also became hard currency for model companies to tell financing stories.

In the eyes of the above-mentioned employees, Wang Xiaochuan's goal of making "super applications" is very grand, but at the product level, "all talks are relatively empty" - Bai Xiaoying's function has been iterated several versions in the six months from the establishment of the project to the release. "He (Wang Xiaochuan) will ask some questions, but will not give answers, and new ideas will continue to emerge."

Facts have proved that Bai Xiaoying's form is still far from the "super applications" in Wang Xiaochuan's eyes. In May 2024, at Bai Xiaoying's press conference, Wang Xiaochuan was not positive about this product: "Bai Xiaoying is a transitional product from AI tools to AI partners with limited model capabilities."

Subsequently, this "negative attitude" product did not splash in the market. It is understood that since its launch, Bai Xiaoying's daily active users have never exceeded 5,000.

However, for Baichuan at that time, compared with the failure of the C-end, he had to face more of the bottlenecks in model research and development.

A employee told Intelligent Emergence that after Baichuan 4 achieved a good market response, Wang Xiaochuan shifted part of the focus of model research and development from text to multimodal.

And unlike generative models such as Wensheng Pictures and Wensheng Video, with the support of Wang Xiaochuan, the production and research team's goal has added a new goal, which is to create a "full-modal model" that can cover text, images, videos, and audio.

"The consensus of Baichuan technical team is that 'the future is multimodal', and the future AGI will definitely be a full-modal model." The above-mentioned employee mentioned that after Baichuan 4 reached the top, the technical team believed that Baichuan already had the experience of training high-performance language models, and the next stage should start to migrate the intelligence of the text model to the multimodal model.

But training full-modal models is difficult for a startup. Due to the immaturity of the technical route and the large amount of computing power consumption, many employees mentioned that Baichuan's progress in multimodal has been very slow.

It was not until October 2024, after nearly five months that Baichuan submitted a multimodal answer sheet, opening the full-modal model Baichuan-Omni. But the first generation of Omni has a bad reputation in the technology community. Many developers believe that Omni's basic capabilities in various modes still have a lot of room for improvement, and it has not reached the point where it can be implemented.

If it weren't for the blockade of the progress of model research and development, many people believed that medical care would not be pushed to the forefront by Baichuan so quickly.

"The movement on the model has been slow, and Baichuan also needs to consider the financing story of the next stage." An investor concluded.

The end of the tail is broken, and the medical rescue is saved. August 2024The once marginal medical team has come to Baichuan Intelligent's business and become the bargaining chip for this former star to raise "winter money".

In the business field that emphasizes historical experience, what the outside world can see from the medical field is the difficulty of implementing AI in medical care and the failure of the previous generation of AI manufacturers.

The four AI dragons, who grew up in 2016, once wanted to get a share of the medical field, but they returned without exception.

IFLYTEK, which currently occupies the largest market share in the AI+ medical industry, suffered a loss of 134 million yuan in the first half of 2024, and the amount of losses has been further expanded compared with the same period.

△iFlytek Medical has been making losses for three consecutive years. Source: iFlytek Medical Prospectus‍

A former employee of iFlytek Medical team told Intelligent Emergence that the accuracy of AI diagnosis and treatment systems is very limited, and the hospital is a complex ecosystem. Each department has unique clinical data and medical processes, so it is difficult to create a general system.

Bachuan Intelligent does not have rich experience in AI medical care. In the eyes of many people, choosing medical care is taking risks, which is a manifestation of Baichuan's "disorganized pace".

But in the eyes of many investors who have talked to Wang Xiaochuan, medical care is the concept of differentiation between Baichuan and the other six Xiaohus, and differentiation is precisely the key to raising money. Nowadays, the government and enterprise market has a smart map, the C-end has the dark side of the moon and MiniMax, and medical care is a relatively blank field.

In multiple interviews, Wang Xiaochuan explained to the outside world that the goal of turning to medical care is not contradictory to AGI (general artificial intelligence): medical care is not a vertical scenario, and creating a doctor is equivalent to AGI.

"At present, Baichuan needs to consider the opponents that Baichuan needs to consider in the medical field and has its own base model. First, Zhishui, which has a lot of government and enterprise resources, and second, iFLYTEK, which currently has the largest medical market share." An employee of Baichuan's medical team told "Intelligent Emergence".

He mentioned that Baichuan's medical treatment has seized a "window period": "Zhipu has not yet finished doing medical treatment, and iFlytek's model capabilities are not as good as Baichuan. It is purely a domestic model supplier and the technology is still in the first echelon, only Baichuan is left."

But Baichuan has to face the problem of entering the medical field. An employee gave an example of "Intelligent Emergence". Baichuan's internal evaluation showed that if it is based only on existing literature and databases, the current generation of AI cases can reach 70-80 points, but the benchmark for the implementation of related projects is 90 points.

The best solution at present is to add labels and alignment from professional doctors. At the end of 2024, Baichuan invested in the medical data service provider "Pediatric Fang" and also established a medical product department with more than 30 doctors.

At the annual meeting, Wang Xiaochuan specially introduced two medical experts hired from overseas and Hong Kong with high salaries, and they were responsible for leading the professional medical team.

"Algorithm engineers are second to train medical models, and their most important role is professional medical staff. "The above-mentioned employee mentioned that whether it is to record a non-structured diagnosis and treatment dialogue to form structured cases or to provide feedback on the cases and diagnosis and treatment results of AI output, professional doctors are needed.

As of New Year's Day 2025, in several internal meetings, Wang Xiaochuan made a very optimistic estimate of 2025 revenue. An employee remembered that he said at the meeting that the 2025 goal is to achieve the performance to the listing threshold (1 billion yuan) and that if you get the Green Channel (the list of green channels) you can go public at any time.

But soon, the entire AI industry ushered in uncertain factors of pattern changes: On January 20, 2025, DeepSeek released a high-performance inference model R1 and opened the source.

The dark horse turned the table, forcing AI manufacturers to concentrate their investment It comes from its own advantageous business. In order to focus on the differentiated advantage of medical care, the guillotine pulled by R1 fell to the B-end group first.

"Smart Emergency" exclusively reported that on February 19, Baichuan adjusted the PE (prompt word engineering) team affiliated to the B-end group to the production and research group. On March 3, the B-end group was laid off. Later, the PE group was also laid off.

But Baichuan's first-mover advantage in the medical field can be maintained, it is difficult to give an answer internally.

Several Baichuan Intelligent Employees told "Smart Emergency" that since last year, many hospital departments have indicators for AI implementation, and customers are willing to sign contracts, "first is for their own political achievements, second is for the ability to issue papers related to AI medical care, and finally for cooperation, to get rebates for subsequent medical product sales. ”

This means that medical signing under the AI ​​trend is likely to become a one-time transaction. “If there is no actual value to medical institutions, signing a contract is unsustainable. "Employee said.

In March 2025, with Huawei announcing the formation of a medical corps, Baichuan was also thinking about how to fight.

"Smart Emergence" learned that Baichuan is discussing how to combine medical care with "Bai Xiaoying" to develop C-end medical products. "Baichuan can establish differentiated advantages in C-end medical products, which can bypass Huawei. "An employee said.

In order to maintain the technical advantages of the medical model level, Baichuan has also recently increased the introduction of professional talents in the medical field, responsible for the data strategy of medical models, feedback reinforcement learning, etc.

A employee told Intelligent Adventure: "At present, Baichuan still has the strongest medical model, and now the company should think of ways to continue this advantage. ”