After three years of intense competition, characterized by a cycle of 'securing finance, burning through cash, and seeking refinance,' the initial phase of China's large model race has concluded, and the true knockout stage has officially commenced.

On December 19, Beijing Zhipu Huazhang Technology Co., Ltd. officially unveiled its prospectus, marking its bid for an IPO on the Hong Kong Stock Exchange. Almost simultaneously, Shanghai MiniMax Technology Co., Ltd. (MiniMax) also released its prospectus to the Hong Kong Stock Exchange. Suddenly, the competition to become the 'first large model stock' intensified, with suspense reaching its peak.

However, not all participants can secure a prime spot in this capital extravaganza.

01.AI, once a highly anticipated contender, has yet to announce a new round of financing or a clear listing strategy. While its peers have secured tickets to the secondary market and accelerated their pace, 01.AI appears to have fallen silent. In the cash-intensive race of large models, the absence of sustained funding means its window for survival is rapidly closing.

This is not merely a race to the IPO finish line but an ultimate test of business models, technological barriers, and capital resilience.

The Divergence Among the 'AI Six Little Dragons'

Currently, the 'AI Six Little Dragons'—Zhipu, MiniMax, Yuezhi'anmian, Jieyuexingchen, Baichuan Intelligence, and 01.AI—are transitioning from technological idealism to commercial reality, with divergence becoming inevitable.

Over the past three years of intense competition, the AI Six Little Dragons started on equal footing, armed with top talent, cutting-edge technology, and abundant capital. Entering 2025, amid a retreat of capital, soaring computational costs, and encroachment by tech giants, they no longer synchronize in their commercialization efforts, financing pace, or strategic direction.

Representatives of the 'deep divers'—Zhipu, MiniMax, Yuezhi'anmian, and Jieyuexingchen—continue to bet on general-purpose large models but have taken divergent paths.

Zhipu AI has constructed a significant business-to-business (B2B) moat with its robust GLM series models and deep strategic deployment in government, finance, energy, and other sectors. It has become the frontrunner closest to the capital market by being the first to file with the Hong Kong Stock Exchange. MiniMax has adopted a more market-oriented approach, with its ABAB series models excelling in multimodal capabilities and reaching millions of users through consumer-facing (C-end) products like voice assistants and content generation tools, simultaneously launching its IPO process.

Yuezhi'anmian, following the stunning debut of Kimi, remains committed to general model research but is gradually shifting its focus to enterprise services and API commercialization, seeking a balance between technological frontiers and cash flow. Jieyuexingchen has not abandoned foundational model research but is betting heavily on 'intelligent terminal agents,' achieving initial breakthroughs in scenarios like in-vehicle systems, robots, and AR glasses.

Representatives of the 'transformers'—Baichuan Intelligence and 01.AI—have adopted more pragmatic strategic shifts, voluntarily abandoning the arms race for ultra-large general models and instead diving into vertical sectors.

Baichuan Intelligence has decisively contracted its general business to fully focus on the medical AI track. Its AI pediatrician and AI general practitioner assistant, deployed in multiple top-tier hospitals, have initially proven viable in commercial loops from technology to clinical value through functions like diagnostic assistance and medical record generation. This pivot, while sacrificing the allure of general AI, has gained clear revenue streams and policy support.

01.AI has also ceased its blind pursuit of trillion-parameter models. Instead, it is collaborating with open-source ecosystems like DeepSeek to focus on lightweight, privatizable industry solutions for enterprises. This pragmatic strategy, though lowering technological ambition, has built differentiated delivery capabilities and client trust in data-sensitive fields like finance and manufacturing.

Whether adhering to general-purpose models or pivoting to verticals, the success criteria for the AI Six Little Dragons have quietly shifted. Parameter scale and benchmark rankings are no longer the sole metrics of value; the ultimate test lies in building sustainable business models and irreplaceable industrial value in real-world scenarios.

A Wave of Skepticism

For 01.AI, the AI large model company founded by Li Kaifu, 2025 promises to be a year of intense pressure.

This year marks its high-profile strategic pivot: abandoning trillion-parameter ultra-large model training to fully bet on enterprise-grade lightweight large models and industry agents. Previously, Li Kaifu stated, '01.AI has temporarily abandoned pre-training ultra-large foundational models with over a trillion parameters. The team now focuses on lightweight, high-performance industrial large models and industry applications.'

This transformation was not a sudden whim but an inevitable choice under multiple pressures.

On one hand, large model training costs have approached the limits of what startups can bear. As a non-listed company, 01.AI lacks cloud business revenue or sustained financing support; continuing to invest in hundred-billion- or trillion-parameter models would be 'slow suicide.'

According to the '2024 Artificial Intelligence Index Report,' the training cost of the Transformer model was approximately $900 in 2017; by 2023, OpenAI's GPT-4 and Google's Gemini Ultra cost roughly $78 million and nearly $200 million, respectively.

On the other hand, the consumer-facing (C-end) market is dominated by tech giants. Its early consumer-facing AI apps, PopAi and Mona, attracted initial attention but struggled to break through user retention and commercialization bottlenecks. Under siege by platform ecosystems like Tencent, Douyin, and Alibaba, independent C-end AI products face near-insurmountable challenges.

Today, 01.AI has launched the Wanzhix Platform and 'Super Employee' Agent, focusing on vertical sectors like energy, gaming, and law, emphasizing privatized deployment, data security, and scenario adaptation. According to official disclosures, it has secured deep collaborations with multiple industry leaders, initially validating the feasibility of its new path.

Rather than a proactive strategic upgrade, this was a forced survival move. Abandoning general-purpose models may save hundreds of millions in spending but also means exiting the competition for next-gen AI infrastructure.

Critically, the strategic adjustment's costs extend beyond business contraction. In 2025, 01.AI faced an unprecedented talent exodus, triggering waves of market skepticism.

Currently, 01.AI has seen seven executives depart, including co-founder Gu Xuemei, Chief Technical Expert Nie Pengcheng, Algorithm VP Huang Wenhao, Technology VP Dai Zonghong, and Chief Architect Pan Xin—five key technical talents. In the AI industry, where technology is paramount, such a massive core talent outflow raises concerns.

However, crisis breeds opportunity. 01.AI has welcomed a new team. In late October, co-founder Shen Pengfei officially took the stage to oversee domestic business-to-business (B2B) and business-to-government (B2G) market expansion; Dr. Zhao Binqiang and Dr. Ning Ning were promoted to VPs, with the former focusing on model and technology system construction and the latter leading international business expansion and AI consulting implementation.

2025 marks 01.AI's true trial of maturity. Executive departures and business contraction have amplified uncertainty but also shed unrealistic burdens. The key question now is not whether it can build large models but whether it can create products customers are willing to pay for.

The Hard Battle Begins After the Glory Fades

After shedding their early hype, the AI Six Little Dragons face a critical transition from technological pursuit to commercialization and profitability. The core challenge lies in validating technical feasibility while building sustainable business models.

01.AI's 'All in B2B' strategic vision remains compelling and aligns with the global trend of large models shifting from technological fervor to commercialization. Amid accelerating digitalization and enterprises' cost-cutting demands, deepening vertical industries and creating standardized products offer a clear and promising path.

However, the market only believes in results. Among the 'Six Little Dragons,' Zhipu's ability to lead the IPO race stems not just from its technological prowess but from its early establishment of a stable revenue structure across 'government + state-owned enterprises + finance.' MiniMax's appeal to Hong Kong investors relies not solely on model capabilities but on its C-end products' user scale and monetization potential.

In contrast, while 01.AI has taken a crucial step, 'B2B implementation' is just the starting point.

Competition in the B2B market transcends technological prowess; it tests industry understanding, resource integration, and service delivery capabilities. The true test for 01.AI lies in whether it can 'enter enterprises,' deeply integrate technological solutions into business workflows, address real pain points, and create replicable, scalable commercial value.

In summary, the entrepreneurial journey of AI large models has never been a sprint but a marathon through cycles. In 2025, as the hype fades, the Six Little Dragons have finally stepped from the spotlight into the muddy real-world battlefield...