After more than a year out of the public eye, Wang Xiaochuan has once again affirmed Baichuan Intelligence's (hereinafter referred to as Baichuan) determination to deeply cultivate the medical track.

In April 2025, Wang Xiaochuan reflected in an all-hands letter that Baichuan had overextended its efforts in the past two years and would subsequently streamline its focus and bet heavily on medical AI (Artificial Intelligence).

"Reconnecting with everyone; I was heavily criticized before," Wang Xiaochuan said with a smile to media reporters, including from National Business Daily, on the afternoon of January 13th. "From day one, we wanted to work on healthcare, but after getting swept up, we did many things outside of medicine. Now we've learned our lesson; every new hire is interviewed directly to see if they genuinely want to work in healthcare."

The industry's direction is shifting. Recently, OpenAI officially launched its Health capabilities, Claude released technologies related to medical computation and Agents, and leading model companies at home and abroad are intensively entering medical scenarios. Wang Xiaochuan judges that medical care, the crown jewel of large models, has begun to enter the realm of application. "We currently lack the capacity to open multiple fronts, so we've chosen to focus solely on healthcare."

Wang Xiaochuan first attributed the core problems in the healthcare industry to "insufficient supply" and "structural imbalance."

Wang Xiaochuan believes the shortage of quality doctors is a long-standing issue. Precisely because of this supply-side deficiency, the internet era did not truly solve healthcare's fundamental pain points, and the emergence of large models makes AI doctors a possibility.

"People might not have believed it much in 2023, but by 2025, they are starting to get a sense of it." Wang Xiaochuan believes healthcare is the crown jewel of large models, and a large model's medical capabilities must ultimately surpass the level of doctors themselves.

Simultaneously, an inequality between doctors and patients has long existed within the medical system. "Healthcare is one of the few industries where the beneficiary and the decision-maker are separate. The patient is the beneficiary; the doctor is the decision-maker." Wang Xiaochuan believes that under the current system, patients often passively accept conclusions, finding it difficult to truly understand the diagnostic logic, differences in treatment options, and risk trade-offs.

Wang Xiaochuan does not subscribe to the dichotomy of AI replacing doctors or AI merely assisting doctors. "What's more important is the transfer of power between doctor and patient, where doctors gradually cede power to patients." He believes the value of AI lies in bridging the information gap between doctors and patients, enabling patients to consult doctors with genuine clarity and understanding.

In Wang Xiaochuan's view, as AI capabilities improve, the gradual transfer of decision-making power to patients is a trend. However, he emphasizes this is not about "taking a slice of the doctor's pie," but rather a structural change in the medical relationship. "A doctor might propose two plans, one conservative, one aggressive. Or three different doctors might suggest three different plans—which one should you choose? Our (AI) doctor is sufficiently powerful to supplement all kinds of information and provide clear explanations. The relationship between patient and doctor will change in this era."

On a macro level, he also pointed out a key difference between Chinese and foreign healthcare systems. Some developed countries have long implemented a family doctor system, with most medical activities occurring at the primary care level. In China, however, patients tend to go directly to top-tier tertiary hospitals, leading to a continuously increasing burden on these institutions.

"A major trend going forward will be a change in the (healthcare) scenario." Wang Xiaochuan believes that as AI integrates into personal lives, many minor illnesses and daily health issues will no longer rely on primary care institutions but will shift to the home environment. "The home will become the front line for one's own healthcare activities."

Under this logic, strengthening primary care no longer refers solely to community hospitals but includes the overall enhancement of diagnostic and treatment capabilities for the home.

On the issue of technical路线 (path), Wang Xiaochuan's stance is equally clear.

Addressing the mainstream view that "multimodal is the main battlefield," Wang Xiaochuan explicitly stated he has never agreed with this perspective from day one. "We mentioned back in 2023 that language is our central axis."

Wang Xiaochuan believes the true震撼 (impact) of ChatGPT lies in AI mastering intelligence, specifically the ability to abstract concrete problems. "Symbols are the core; analogy is intelligence."

Wang Xiaochuan categorizes natural language, mathematical language, and code language uniformly as formal languages. "Previous mathematicians have said that mathematics is about depicting different things with the same concepts, so symbols and language are the central axis." He believes that, to date, evaluating the strength of a company's model capability is fundamentally based on symbols; high usability does not equate to high intelligence.

Specifically in the medical field, Wang Xiaochuan pointed out that a vast number of problems are essentially decision-making problems. "If it were just an image problem, then reading the scan would be enough."

Based on his judgment, future hospital image interpretation will likely be handled by specialized small models. Once these results are symbolized, they will be handed over to language models for reasoning tasks. "Perception models and cognition models need to combine. Recently, there's even a non-contrast CT model for pancreatic cancer; compared to intelligence models, these are like 'small leaves hanging on a tree,' not the main battlefield."

This is also why he repeatedly emphasizes that image recognition does not equal intelligence. "We stress that the technological红利 (dividend) occurs in intelligence, not in image recognition. Image recognition does not represent intelligence; it's just the work of a test-taker."

Regarding the issue of data, Wang Xiaochuan's views are equally contentious. "People always make a big deal about data. Understanding the matter and having the funds to do it are the key factors. There are over 500 vertical medical models on the market; I believe they don't even know what a model truly is, so don't be influenced by them."

He does not agree with the view that high-quality in-hospital data determines a model's upper limit. Wang Xiaochuan believes intelligence is the fundamental issue, not a backup plan to compensate for data shortcomings. "Data is important, but there are no success stories in the large model industry; everyone has got it wrong. Everyone talks about how important data is, but it's just a concept being passed around; no one has successfully done it based solely on that."

In his view, language models rely more on a transferable knowledge system, derived more from academic papers than from raw data itself. "Transforming the hospital's high-quality data into knowledge, and then extracting this information from papers is the key. This is different from the previous generation's image-based methods. Images aren't directly transferable; knowledge is transferable."

He further pointed out that the real core is not the data source, but the evaluation system and training methodology. "Having doctors help us produce an evaluation system that simulates patients is a different matter from training models with data."

On scenario selection, Wang Xiaochuan repeatedly emphasized one judgment: the massive future growth for medical large models lies outside the hospital, not within it.

"The hospital is more a place for surgeries and IV drips... the (in-hospital) space itself is limited." In his view, introducing AI into medical record systems and quality control processes is not the future mainstream direction, but rather an inefficient approach, like "hitting the target indirectly."

"The most important thing is to help patients directly." Wang Xiaochuan believes that the previous generation of medical AI products were doctor-centric service models, which are far from the essence of AI. "We always say the patient is the primary person responsible for their own health, but AI has been helping doctors, not patients. It's always been an indirect approach—is that right?"

Baichuan has chosen a clearer To-C (consumer-facing) path. Wang Xiaochuan revealed that Baichuan will enter the market in the first half of this year, launching two products.

On commercialization, he does not shy away from the issue of charging fees. "As long as it can help patients generate value in辅助决策 (assisted decision-making), it can charge a fee." Whether charging patients directly or integrating medical resources and pharmaceuticals/equipment through service packages, he sees neither as particularly difficult challenges.

On regulatory boundaries, Wang Xiaochuan's attitude is cautious. "We will not cross red lines; we will not perform diagnoses ourselves or prescribe medication, but we can do辅助工作 (assistive work)." Baichuan's positioning is to help patients understand information, weigh options, and participate in decision-making, not to replace doctors in drawing conclusions.

When discussing Baichuan's moat, Wang Xiaochuan summarized it into three points: having a model that leads by a generation,切入 (entering) high-value, serious scenarios, and adhering to an innovation rhythm different from that of large tech companies. "Major innovations come from small companies; minor innovations come from large companies. We must focus on what we believe are high-value endeavors. Consensus is not our priority for breakthroughs, whereas large companies focus more on consensus; our roadmaps and product forms are different."

When asked about overseas expansion, Wang Xiaochuan stated plainly that Baichuan will definitely go global. "A medical company that cannot go global is not a good company."

Additionally, when asked about his views on the recent IPOs of智谱 (Zhipu) and MiniMax, Wang Xiaochuan indicated that the two listed companies capitalized on the technological红利 (dividend) of general-purpose models and the foundation of national support for technology. AI healthcare is also a significant force in the current large model competition, but compared to general models, it requires a longer timeframe. Therefore, Baichuan will also take the path to an IPO in the future.

"It's just that the timing will be a bit later. We hope our business model can be more optimal, including the structure between revenue and costs, so we believe we still need one or two years of preparation time," Wang Xiaochuan said.

Finally, Wang Xiaochuan mentioned his original motivation for entering this field. He recalled his early studies in weather forecasting and genomics, and his long-standing curiosity about the complexity and regularity of life.

"Life is more complex than weather forecasting—how can there be underlying规律 (rules/laws)? So I spent time researching, always wanting to find the mathematical model behind it. Now we need data-driven approaches, and language models can solve intelligence problems. This was my initial motivation," Wang Xiaochuan said.