Our readers are relentless :) Nearly 40% voted for an article about a GenAI unicorn this Thanksgiving weekend, and I’m grateful for the enthusiasm. Today, we’re diving into the story of Baichuan Intelligence (百川智能) – one of China's most innovative companies, valued at $2.8 billion.

Led by Wang Xiaochuan, who believes we’ve moved from the information era to the intelligence era – and that mathematics, as the ultimate truth, will guide humanity into a symbiotic future – Baichuan isn’t focused on building large models simply for their size. Instead, the company blends long-term strategy with practical innovation, developing "super models" and "super applications." With a portfolio of 12 large models, including the open-source Baichuan-7B and proprietary Baichuan 4, this Beijing-based startup is reshaping AI applications across healthcare, education, and finance.

However, investors remain cautious: Wang’s unconventional approach challenges norms, sparking debates about its sustainability in an intensely competitive market.

What sets Baichuan apart in China’s crowded AI landscape? What drives the vision of math prodigy Wang Xiaochuan, and what does this vision mean for the future of Baichuan – and for its investors? Let’s explore.

In today’s episode:

Two years ago, on November 30, 2022, the release of ChatGPT changed the world forever. For Wang Xiaochuan, it marked a personal "before and after" moment. That day, he felt a mix of emotions: awe at the United States' technological lead, a sense of urgency to catch up, and a spark of excitement to begin something transformative. Questions about strategy, resources, and talent started forming in his mind.

Wang’s pivot to AI came after stepping down as CEO of Sogou in 2021 following its acquisition by Tencent. Financially secure, he initially ventured into medicine and life sciences. But the rise of large AI models inspired him to merge these interests into a singular mission: using AI to tackle systemic challenges like the global shortage of skilled doctors.

This vision wasn’t about incremental changes – it was about rethinking how AI could address underserved needs, starting with healthcare.

In February 2023, he posted on Weibo:

He was already “pregnant” with Baichuan.

It’s a bit of a puzzle who exactly the co-founders of Baichuan are. Officially, the website names only Wang Xiaochuan as the founder, and that’s the story it’s sticking to. But LinkedIn throws a twist into the mix, listing Ke Jiao as a co-founder since April 2023 – a Tsinghua graduate with a modest 31 connections but an official-sounding title. Meanwhile, various media sources claim that in March 2023, Wang announced the launch of Baichuan Intelligence with Ru Liyun, his long-time ally from the Sogou days, as a co-founder.

Oddly, Ru’s LinkedIn points to another venture they co-founded in 2022, Wuji Zhikang (The phrase 五季智康 can be translated as "Five Seasons Wisdom and Health" or "Five Seasons Intelligence and Wellness) focusing on healthcare and AI. So, what exactly they co-founded together?

In reality, it’s less about separate companies and more about a multi-layered strategy. Wuji Zhikang, founded in 2022, laid the groundwork with its focus on health consulting and AI software. Baichuan Intelligence expanded on this in 2023, with a stronger emphasis on AI application software and model development.

It seems all these ventures are part of the same ecosystem, unified under Wang Xiaochuan’s ambition to build a "Chinese OpenAI" and create an "AI Doctor" that bridges cutting-edge AI with transformative healthcare applications.

To understand the Chinese AI Tigers and their potential for success, one must first recognize their deep connection to Tsinghua University. All the founders share a common thread: they studied at Tsinghua. Some are now professors there, and all operate within walking distance of the campus.

The key difference among them lies in speed: Zhipu AI was founded in 2019, while both Moonshot and Baichuan Intelligence emerged in March 2023. But Moonshot is led by Yang Zhilin, a young AI researcher celebrated for his groundbreaking work. Baichuan, in contrast, is helmed by Wang Xiaochuan, once hailed as a prodigy and even called a genius in his youth. However, Wang’s career followed a different path – he spent 20 years building Sogou, a search engine that, despite being acquired by Tencent, never achieved tech giant status. According to some Chinese media, because of that his mother considers him a disappointment.

At Tsinghua, where excellence is the norm, Wang’s pace seems slow. Progress there is relentless: “If you have 5 minutes of free time and don’t make any progress, you’ll start to feel depressed.” Tackling the hardest problems with unwavering focus is a hallmark of the “Tsinghua team” driving the development of large-scale models. First, Baichuan achieved a notable speed of development, releasing 12 large models in one year, but then it slowed down.

Wang Xiaochuan sees humanity at a pivotal moment: we’ve moved beyond the Information Era and entered the Intelligent Era, where AI has mastered language, transforming it into mathematical models. As Wang puts it, "this is the biggest advance in human civilization since Newton transformed physics into mathematics." But this is only the beginning. He envisions the next milestone as the Symbiotic Era, where AI evolves from a tool to a collaborator, bridging intelligence with the complexities of life sciences and consciousness.

Baichuan’s approach reflects this philosophy. The company focuses not on building large models for their own sake but on creating "super models" and "super applications" that address high-impact, knowledge-intensive challenges. Wang calls healthcare the “crown jewel” of big models, aiming to create AI doctors capable of operating at the level of top-tier physicians. These applications, he argues, are not mere tools for niche tasks but steps toward integrating AI into broader human systems.

A key aspect of this transition is the shift from fast thinking to slow thinking. Wang emphasizes that future AI systems must take time to reason and generate meaningful insights. "Machines are learning to think slowly," he notes, highlighting how deeper, structured reasoning—though computationally demanding – produces richer, more thoughtful outputs. This marks a departure from the rapid, surface-level decision-making of earlier AI systems.

Baichuan’s work in language modeling exemplifies this structured intelligence. For Wang, language is foundational – a gateway to modeling knowledge, cognition, and communication mathematically. This belief drives Baichuan’s dual focus on advancing AI systems that not only process information but also collaborate creatively with humans to solve complex problems.

The company aims to transcend national boundaries: “We do not want to be just ‘China’s OpenAI,’ but to establish a world-class vision and technical quality.” Symbolizing this vision, Wang describes Baichuan’s name: “Baichuan means many rivers converging into a powerful intelligent system, symbolizing endless wisdom and collaboration.”

In August 2024, Baichuan Intelligence finally announced that it was all in on healthcare. On August 28, 2024, Baichuan Intelligence signed a strategic cooperation agreement with Beijing Children’s Hospital. Wang explained the significance of this collaboration:
“We believe that within three years, we can develop an AI pediatric doctor with the expertise of a chief physician at a top-tier hospital. This would be equivalent to creating one million such doctors, enough to serve clinics at the township level nationwide.”

China’s biggest e-commerce and cloud services operator has emerged as one of the most prolific backers of Chinese AI start-ups, with stakes in all four AI tigers, including Beijing-based Zhipu AI and Shanghai-based MiniMax.

Alibaba is actively developing its own large language models (LLMs), including its proprietary Tongyi Qianwen series and the open-source version called Qwen. Just last week, on November 27, 2024, Alibaba released a preview of Qwen QwQ /kwju:/ – an open model designed to advance AI reasoning capabilities. Twitter burst into praise. And we can’t help but wonder, with Baichuan’s passion for reasoning and slow thinking, how much Wang might be biting his elbows and what he’ll come up with next.

Now, to what they’ve presented so far:

Overview:
Baichuan-Omni is Baichuan’s first open-source 7-billion-parameter Multimodal Large Language Model (MLLM) capable of processing text, images, videos, and audio simultaneously. Released in October 2024, it leverages cutting-edge innovations in multimodal alignment and multitask fine-tuning to deliver state-of-the-art performance across modalities, setting new benchmarks in open-source AI.

Key Features:

Performance Highlights:

Open-Source Impact:
Baichuan-Omni provides its model, training code, and evaluation scripts openly, fostering collaborative research and innovation in multimodal AI.

Future Directions:
The team aims to enhance capabilities in environmental sound recognition, extend context handling for longer video/audio sequences, and integrate TTS systems for richer interactions.

Why It Matters:
Baichuan-Omni bridges the gap between proprietary and open-source multimodal AI, democratizing access to cutting-edge technology. Its robust design and high performance position it as a key driver for both academic research and enterprise applications.

It’s interesting, by the way, how much attention Wang pays to academic support. On Baichuan’s website, there is a whole section dedicated to it:

As for the future of their MLLM, the team isn’t done yet:

That’s not exactly clear. For enterprises, Baichuan offers a cost-efficient AI deployment solution with two models: Baichuan4-Turbo and Baichuan4-Air. Turbo excels in multilingual tasks, reducing inference costs to 15% of its predecessor, while Air, using MoE architecture, cuts costs to 0.98 yuan per million tokens. Both models feature high availability and optimized performance for enterprises.

But in general, that’s not what Wang is concerned about. People close to him say: “He won't follow Sora, he won't fight for long texts, and he won't participate in price wars. Everyone is anxious about commercialization, but he doesn't care.”

While the industry is abuzz with discussions about commercialization, it’s not his current priority. He entrusted this aspect to Lianchuang Ru Liyun, emphasizing that the goal is simply to avoid losses. With sufficient funds in the account, responsible spending, and steady cash flow, he believes Baichuan has no immediate financial concerns. Moreover, since competitors aren’t profiting either, there’s little urgency for Baichuan to focus on revenue generation.

That’s Wang’s reasoning: “In an age of increasing aging, people's demand for medical care has increased, which has led to a trillion-dollar market. If you look further, the big model is a mathematical model of language, and the medical big model is a mathematical model of life. "The world of life is a bigger thing than the physical world."

Wang Xiaochuan has a profound vision of AGI's transformative impact on society. He sees AGI not just as a productivity tool, but as a force that will fundamentally reshape human civilization.

He believes and is most fascinated by the fact that language becoming mathematics is this generation's biggest advancement, enabling knowledge, thinking, and communication to be transformed into mathematical problems. He envisions AGI leading to a "symbiotic era" where AI will tackle complex challenges in life sciences and consciousness through mathematical modeling.

Looking ahead, Wang sees machines gaining the ability to autonomously write and execute code within 5-10 years. He envisions a future where humans and AI collaborate closely, forming a civilization where intelligence is shared and mutually enhanced. Rather than discovering new needs, he believes AI will primarily address existing demands with unprecedented efficiency, particularly in critical areas like medicine and education.

The case of Baichuan Intelligence is extremely interesting. Under Wang Xiaochuan’s leadership, the company embodies a balance between ambition and pragmatism, today’s expectations and tomorrow’s possibilities. Wang is not chasing trends; instead, he is building a foundation for what he calls the "Symbiotic Era" of human-AI collaboration.

Among the AI tigers we’ve covered, Wang stands out as the most idealistic, firmly rooted in his belief that mathematics is the ultimate truth and a universal framework for AI’s evolution. Baichuan’s approach is unconventional: prioritizing slow, deliberate thinking over rapid execution, and long-term impact over immediate commercialization. Wang envisions a future where AI not only solves productivity challenges but addresses deeper, existential problems.

This vision is both a bold gamble and a significant risk. Investors have raised concerns about Baichuan’s slower pace and its divergence from conventional strategies. Yet, Wang remains resolute, channeling his expertise and mathematical genius into Baichuan’s mission. For the first time, he holds complete decision-making authority, using it to chart an independent course for the company.

One thing is certain: Baichuan is forging its own path in the AI landscape, not following anyone else’s lead. Remember the company Wang co-founded with Ru Liyun—Wuji Zhikang? Consider it the stepping stone that laid the groundwork for Baichuan’s vision. Together, these ventures reflect Wang’s broader strategy: an interconnected ecosystem that blends AI innovation with real-world applications, particularly in healthcare.

Thank you for reading!