In the global fintech landscape, quantitative hedge funds represent a group of stealthy yet formidable "hunters." They rely on mathematical models and massive datasets to build formidable technological moats through millisecond-level trading strategies. In China, leading quantitative private funds such as High-Flyer Quant, Jiukun Investment, and KD Investments not only manage vast assets exceeding 60 billion yuan but are also pushing the boundaries of financial engineering precision through relentless algorithm iteration and computing power competition.

With the rapid advancement of artificial intelligence, these three firms, standing at the pinnacle of both capital and technology, have simultaneously turned their attention to a broader frontier: large language models. Research indicates that these quant giants are not merely engaging in a simple, homogeneous crossover but are pursuing a strategic breakout through divergent paths. While all three operate under the "large model" banner, their underlying evolutionary logics differ significantly.

One seeks to replicate the breadth of human intelligence, aiming directly for the ultimate form of Artificial General Intelligence (AGI). Another chooses to delve deeply into vertical domains, striving to become a super-specialist in specific industries. A third transcends the boundary between digital and physical realms, exploring how to unlock the secrets of unknown scientific frontiers. Leveraging their acute data intuition, ability to model complex systems, and inherent willingness to heavily invest in computational resources, these quant firms imbue these three distinct paths with immense potential.

This "Large Model Evolution" initiated by the three giants signals a unique, diversified path for the next phase of China's AI industry implementation. The ultimate value of artificial intelligence may lie in whether it aims to build an omniscient and omnipotent "deity" or cultivate a group of highly skilled "artisans," each excelling in their own domain. The answer lies within the choices made by these three institutions.

**DeepSeek: Aiming for the "Versatile Genius"** Within the quantitative private fund sphere, High-Flyer Quant has always been a unique presence, gaining prominence in the A-share market since 2018, with its net value curve later becoming a bellwether for quantitative trading. In May 2023, DeepSeek was formally established, spearheaded by High-Flyer's actual controller, Liang Wenfeng, incubated within the High-Flyer Quant system, and receiving continuous funding and computational support. DeepSeek has a singular goal: AGI.

Today, the term "AI" has become overly generalized, often used as a catch-all label in the tech world. However, the AGI pursued by DeepSeek is fundamentally different from the prevalent "weak AI." AGI aims to cultivate a "versatile genius." It is not content with excelling at individual tasks but seeks to endow machines with general reasoning abilities, autonomous learning capabilities, and cross-domain transfer skills akin to humans. Essentially, AGI is a "generalist" that understands the underlying logic of the world; it could analyze stock market charts one day, derive physics formulas the next, and autonomously write complex software systems thereafter.

DeepSeek's choice signifies a move beyond being mere "hunters" in the securities market towards aspiring to become "creators" in the intelligent era. This techno-idealist ethos is vividly reflected in their recent talent recruitment efforts. DeepSeek's latest job postings describe core talent needs with distinct emphasis: seeking individuals with innate curiosity extending to exploring the underlying mechanisms of complex systems, dissatisfaction with passive execution, independent thinking about AGI's technical path, and a drive to explore high-value agent application scenarios. This translates to a rejection of mere "tool operators" and a search for "essential explorers" who seek to understand the "why" behind problems.

Research also notes that over the past year, the DeepSeek team has published multiple papers on innovative underlying architectures, with Liang Wenfeng as a core author. These papers address key areas like long-context sparse attention, residual connection constraints for training stability, and conditional memory mechanisms separating knowledge from reasoning, tackling core foundational architecture innovations for large language models to make training more stable, long-context processing faster, and reasoning more computationally efficient.

**Jiukun Investment: Tackling Vertical Applications for Large Models** Jiukun Investment is a benchmark institution among China's first generation of quantitative private funds, its development history mirroring that of the country's quantitative investment industry. As early as the end of 2017, its assets under management surpassed 50 billion yuan, making it the largest quantitative firm in China at a time when most peers managed between 10-20 billion yuan. Entering 2018, Jiukun continued its momentum, rapidly expanding its scale and solidifying an industry landscape often described as "Jiukun in the north, High-Flyer in the south," with rumors of its scale breaking the 100-billion-yuan mark circulating in recent years.

While advancing rapidly in quantitative investment, Jiukun did not rest on its laurels. Recognizing the transformative potential of AI for various sectors, Jiukun formally entered the large model arena in 2025, establishing the "ZhiZhi Innovation Research Institute." However, unlike DeepSeek's focus on building a "versatile general brain," Jiukun chose a different path: it倾向于 trains "expert-type" vertical large models, not pursuing omniscience but focusing on specific deep-water areas like code writing and medical diagnosis.

On the first day of 2026, the institute announced the open-source release of a new code-specialized large language model. This model functions as an AI assistant capable of reading, writing, and modifying code, helping programmers with automated programming, debugging, and code explanation. Analysis of the institute's website reveals three core research thrusts: First, enhancing model efficiency for complex data processing, making models more cost-effective, better at understanding symmetrical relationships in data, and linearly scalable—akin to building a more powerful "CPU." Second, evolving models from mere code-writing "tools" into autonomous "agents" capable of thinking, planning, and executing tasks. Third, developing multimodal large language models for general medical visual reasoning, enabling models not just to "see" 2D and 3D medical images but to "understand" complex information like surgical videos, acting as a "super assistant" to doctors.

Perhaps the answer Jiukun is proposing is that AGI evolution should not be a castle in the sky. This firm is attempting to use open-source code models as a lever to catalyze intelligent transformation across more vertical fields, ultimately laying the groundwork for the scalable implementation of AGI.

**KD Investments: Targeting the "Scientific Research Super-Person"** In 2023, against a backdrop of overall A-share market weakness and extreme fundraising difficulty, KD Investments defied the trend by crossing the 100-billion-yuan threshold, adding over 10 billion yuan in new assets and becoming the "fundraising champion" of China's securities private fund circle that year. This achievement was largely attributable to its outstanding performance during the preceding bear market.

Bolstered by substantial capital strength and a keen sense for technology, KD Investments officially entered the large model arena in 2025. Observations indicate that KD's large model positioning differs not only from DeepSeek's grand AGI narrative but also fundamentally from Jiukun's vertical focus: KD aims to cultivate an "explorer" at the frontier of "unknown" science, directly accelerating humanity's breakthrough in understanding natural laws.

A closer look at KD's Intelligent Learning Lab's official statements reveals their goal: to create a "Super Science and Technology Assistant (ASI for Sci-Tech)." According to its strategic vision, the core logic is clear: "Focus on developing a universal super assistant for science and technology, pursuing technological compound interest and sustained leadership." In simpler terms, the lab wants to build a super assistant dedicated to serving scientific research and technological innovation. KD's communications mention "exploring the upper limits of AI capabilities through industrial-grade R&D methods" and "continuously advancing model capability development and deep application exploration for science and technology fields."

So, what kind of personnel are needed for this? Recently, KD's key recruitment focus has been on AI Researchers. The official description for this role is specific: "Focus on developing next-generation AI models and algorithms, conducting innovative research on large language models and achieving scalable implementation. Covers four main directions: pre-training, post-training, data & evaluation, and intelligent agent systems."

From KD's布局, they appear to have found a differentiated track: deeply cultivating B2B scientific research and industrial scenarios. This aligns with their quantitative institution's rigorous demands for data and logic while providing their substantial capital with a technologically imaginative long-term outlet. Whether this "Super Science and Technology Assistant" can truly become an indispensable aid for researchers remains to be seen.