"First Listed Company in Decision-Making Large Models: Zhongke Wenge’s Path to Advanced Decision Intelligence"
In its early days, Beijing Zhongke Wenge Technology Co., Ltd. (“Zhongke Wenge”) did not have the clear market positioning it enjoys today.
In the winter of 2018, the team encountered a snowstorm that blocked the roads during a business trip in the north, leaving them stranded on the highway late at night. Such travels were not uncommon at the time. To help customers truly understand what problems artificial intelligence could solve, the team had to visit one client at a time, explaining industry by industry.
When AI and algorithms were mentioned back then, the typical response was “That thing doesn’t seem that relevant to me.”
Seven years later, on June 26, 2026, Zhongke Wenge was listed on the Hong Kong Stock Exchange.
On June 25, in pre-market trading, Zhongke Wenge closed at HK$115.4, up 90.12% from the IPO price of HK$60.7. The Hong Kong public offering was oversubscribed by 5966.78 times; the international placement was oversubscribed by more than 20 times.
On the first day of listing, Zhongke Wenge’s intraday gain reached 87%.
In 2025, Zhongke Wenge’s revenue exceeded 400 million yuan, and the net dollar retention rate (NDR) for existing customers reached 139.5%.
By revenue, Zhongke Wenge ranked first among providers of decision intelligence services driven by large enterprise AI models in China this year, with a market share of 10.2%.
Unlike the early startup days when it was necessary to repeatedly explain the problems AI and algorithms could solve, recently, Zhongke Wenge’s frontline salespeople have clearly felt increasing demand during intensive visits to dozens of small and medium-sized enterprises in Jiangsu and Zhejiang.
“People say they want to buy as soon as they meet us.” For example, in a textile factory, the product can serve as both an AI fault diagnosis assistant and a management decision aid; a foreign trade boss can ask a question in natural language, and the system automatically connects to underlying data and generates a decision report.
This change is underpinned by an eight-year effort that produced a technical framework called DOMA, organizing data, industry ontology, large models, and agents into a complete chain, enabling AI to be embedded in every production decision-making link within enterprises.
According to Zhongke Wenge’s assessment of the next stage of AI, the truly important question is no longer simply “Can AI answer questions?” but rather “Can AI further simulate the world and predict the future?”
This assessment is the basis for Zhongke Wenge’s recent release of the general decision large model Decitron Decision Engine. Co-founder and CEO Luo Yin states, the key for the decision engine is not answering questions but turning the complex world into a system that is computable, simulatable, and verifiable.
The following is the complete story of this “CAS-affiliated AI company” from laboratory to IPO.
Looking for Nails with a Hammer
The story begins in Zhongguancun.
Before 2017, the core members of the Zhongke Wenge founding team were researchers at the Institute of Automation, Chinese Academy of Sciences. During this period, the team gradually formed a clear consensus: the value of AI should not remain only in laboratories and papers, but should enter industrial practice, being transformed into real business and economic value.
In 2016 and 2017, the country successively issued policies encouraging scientific researchers to leave their posts and start businesses. Against this backdrop, PhD-level researchers Wang Lei, Luo Yin, Zeng Dajun, and others jointly founded Zhongke Wenge.
Wang Lei is co-founder and chairman, Luo Yin is co-founder and CEO, Zeng Dajun is co-founder.
“CAS told us, ‘If the startup fails, you can come back.’” Wang Lei recalled to Wallstreetcn, “But then we found we couldn’t go back. The team had left together, the company was so big with so many colleagues, and everyone needed to work here, earning their own income.”
The company was named “Wenge” from the phrase “Wen Xian Ge Er Zhi Ya Yi,” meaning to understand elegance from hearing music. The company’s mission sounds grand: to create advanced business and decision systems for enterprises in the AI era.
Despite having the CAS background, Zhongke Wenge had no customers, no sales team, and hardly anyone could understand what they were talking about when they started.
“Indeed, with our background, many customers trust us. But when it really comes to market competition, whether you’re from CAS or Tsinghua, everyone’s on equal footing,” said Wang Lei. “Ultimately, you have to provide valuable service to customers.”
Zhongke Wenge’s first foothold was the media and communications industry, which, from a capital markets perspective, is not a sexy field.
But Zhongke Wenge had its own judgment: In 2016 and 2017, there was no general large model like today; the technology was based on small to medium-sized data and specialized machine learning. But media and communications had vast, complex, multimodal data. Texts, images, videos, and audio all needed algorithms for recognition and understanding, and there was a clear willingness to pay.
The team knew, they had to first find a “nail” to hammer in, before expanding a larger business map.
Helping large media organizations analyze complex data and assisted decision-making with AI was Zhongke Wenge’s first “nail.”
At that time, the major media institutions faced the challenge of daily uploads of video, audio, text and other diverse formats of interview materials being too complex for effective use and storage.
To address this pain point, Zhongke Wenge annotated, stored, and labeled interview materials like images, audio, and video for semantic retrieval, facilitating later processing and dissemination.
This may sound mundane, but for large media groups processing vast amounts of materials daily, this fixed workflow approach genuinely improved efficiency.
Thus, Zhongke Wenge gradually built up full-chain service capabilities ranging from data management, content analysis, to topic selection and dissemination effectiveness evaluation, winning multiple media clients.
Though the industry had limited imagination, it became the “straw” for Zhongke Wenge during difficult times.
In the toughest year, 2020, Zhongke Wenge relied on this client for relatively stable cash flow, keeping the company alive through the earliest AI commercialization winter.
For Zhongke Wenge, this experience provided not just clients and revenue, but also underlying capabilities later proved invaluable: habits of handling multi-source heterogeneous data, systematic modeling methods for “event-object-relation-impact,” and engineering experience in high-governance scenarios guaranteeing reliability.
With media as a base, Zhongke Wenge began to expand into government, finance, scientific research, and industry.
Wang Lei used an analogy to describe the efficiency of this migration: “Previously, after doing System A, you would have to start from scratch for System B. Now, after doing A, for B, maybe 70% is accumulated; you only need to do 30%. Eventually, after doing many, you’ll find maybe 90% is generic stuff.”
Looking back today, that “nail’s” true importance was allowing the team to develop the ability to handle complex organizational data, understand business rules, and build industry knowledge systems for the first time.
Later, entering government, finance, scientific research, and industry, these capabilities were almost entirely reused.
A Decision That Had To Be Costly
At the end of 2022, ChatGPT burst onto the scene,
The core team at Zhongke Wenge held meetings until midnight, discussing the possible impact. They believed this was not just a product-level innovation, but possibly a migration of fundamental technical paradigms.
“Previously, specialized machine learning solved problems in specific industries, and algorithms worked for domain A, but you’d need to retrain for domain B.”
After ChatGPT, Wang Lei felt that large models changed the entire competitive logic of the AI industry.
From “training a model for each problem” to “building a universal intelligent base capable of understanding the world.” If they missed out on building foundational model capabilities for this round, the company could lose its future eligibility to compete.
At the time, the team faced a major choice: Should they train their own large model?
Training a large model meant multiplying R&D investment; not training meant potentially being eliminated.
“Just as the company was about to make a profit, you’re now spending again on this. Is the money going to come back?” Investors questioned it then.
After repeated weighing, the team judged: “If we don’t do this, maybe in the future Wenge won’t exist. In a few years, Wenge will be eliminated from this track.”
Early 2023, Wang Lei and Luo Yin called on CAS Automation colleagues to discuss the plan and decided to run through the training process themselves.
The financial cost of this “big bet” was clearly reflected in reports.
That year, Zhongke Wenge’s R&D spending reached 180 million yuan; net losses reached 260 million yuan.
The expanded losses meant Zhongke Wenge made up a key piece: foundational model capabilities.
In June the same year, Zhongke Wenge released its fully self-owned “Ya Yi” large model, becoming one of the earliest enterprise AI companies in China with end-to-end foundational model training capability.
Foundational model capability thus became an important factor in Zhongke Wenge’s market competition.
According to Zhongke Wenge, after 2023, during client projects and bidding, the company increasingly sensed the differentiation in enterprise AI competition: Companies without foundational model accumulation face technical boundaries in designing solutions for complex scenarios; model R&D capability alone is not enough, as deep understanding of industry business and decision closed loops is also needed.
Companies need not only model capability, but a systematic capability deeply integrated with business scenarios.
Zhongke Wenge’s model system now comprises two types: Ya Yi, a general large model, and Pan Shi, specialized large models for niche scenarios.
The guiding idea is “integration of general and specialized”—the foundational model addresses basic and general problems; specialized models meet deep needs for vertical sectors.
No single vendor can use one model for everything. Those strong in basic models may not excel in scientific research or code models. Zhongke Wenge believes the model ecosystem will ultimately feature a layered structure: basic models, industry models, and specialist models.
The “Last Mile” of AI Implementation
If judged solely by general model parameter size and computational investment, Zhongke Wenge may struggle to match internet giants and other AI players.
But enterprise AI competition is not just a race in foundational model capability. For companies like Zhongke Wenge, the real key is: Can you combine large model capacity with industry knowledge, business data, and organizational processes to solve complex decision problems in real scenarios?
Many enterprises fail with AI not because the model isn’t smart enough, but because the model doesn’t understand the company’s business rules. Zhongke Wenge proposed the DOMA framework—Data, Ontology, Models, Agents—which sounds academic but is an important methodology developed from years of industry practice.
Data, ontology, models, and agents together form the foundation for enterprise AI operation, with industry ontology as the most critical layer. It allows AI to not only understand language but how enterprises think.
Take investment analysis: If a model is asked to evaluate an investment target, Person A asks using one framework, Person B another.
But if you tell the model beforehand that evaluating a company should focus on financial reports, core team, and technological advancement, then direct the model to perform quantitative and qualitative analysis accordingly, the consistency and accuracy of results are greatly improved.
This “telling the model how to look at data” process is ontology modeling.
The ontology layer structures business objects, their relationships, operating rules, and constraints in an industry, forming an AI-understandable “industry thinking pattern.”
In recent years, the biggest value of large models has been information acquisition and content generation; but enterprises care not only about “the answer,” but “what to do next.”
A Decitron developer told Wallstreetcn that many complex decision problems are difficult because they involve multiple entities, variables, and evolving constraints, and cannot be reduced to one-shot Q&A.
“Real decision problems must be converted into analyzable, decomposable, simulatable structures.” In their view, this is the key difference between decision engines and traditional generative models: the latter excels at generating answers based on prior knowledge, while the former emphasizes ongoing modeling of event relationships, path changes, and result branches.
The model sees data not as a blind person feeling an elephant, but with a clear analytical framework.
Zhongke Wenge’s new DIP (Decision Intelligence Platform) is also a productization of this methodology.
DIP addresses the three most common challenges in enterprise AI implementation: scattered data, complex business, and difficulty advancing action.
In enterprise scenarios, a single order is often linked to a chain of clients, products, inventory, contracts, and more. If AI only reads data but doesn’t understand their interrelations, it can’t participate in genuine business decisions.
DIP’s role is to turn dispersed data from different systems into “business ontology” understandable by AI, then let the model make analysis, judgments, and actions based on business rules.
For example, staff want to know “which Grade A medical industry clients might churn in the next 30 days”—DIP will automatically link client grades, order history, after-sales work orders, etc., forming a complete client operations view. Finally, it produces reports listing at-risk clients, explanations for risk, etc.
This means Zhongke Wenge doesn’t plan to compete with the giants on parameter scale, but aims to win the “last mile” of enterprise AI: enabling models to read business, understand rules, and enter the enterprise’s decision process.
Decitron Decision Engine:
Enabling AI to Simulate Decisions, Not Just Generate Answers
In past years, large models have shown AI’s capabilities in content generation, knowledge Q&A, and efficiency improvements. Zhongke Wenge believes enterprise AI now is moving from “able to answer questions” to “able to aid decisions.”
This is the reason for the recent launch of Decitron Decision Engine.
As a general decision intelligence model for complex open scenarios, Decitron Decision Engine isn’t limited to a single industry or task; it focuses on complex decision problems under uncertainty, multi-path choices, and multi-party games, helping users understand situations, simulate trajectories, compare schemes, and make more valuable judgments.
Its application covers finance, macroeconomics, international affairs, public governance, enterprise strategy, investment research, industry analysis, etc.—a concentrated productization of Zhongke Wenge’s decision intelligence capabilities.
For example, predicting US interest rate changes: Interest rate decisions are determined not by a single data point, but by inflation trends, employment data, economic growth, market expectations, policy signals, international developments, and more.
Facing such highly uncertain complex issues, Decitron analyzes multi-source information and simulates many paths, helping users understand possible directions, key variables, and potential impacts under different scenarios.
This shows the difference between decision intelligence and traditional information retrieval or generative Q&A: the focus is on helping users make judgments in uncertain environments.
In everyday life, Decitron Decision Engine can be used for “college entrance exam voluntary submission.” When scores are released, volunteers planning becomes a crucial choice for students and parents. For this, Decitron provides “volunteer simulation” to assist.
Unlike tools mainly based on existing data and rules that provide relatively certain recommendations, Decitron emphasizes simulating opportunities, risks, and differences behind each choice, helping users make judgments based on understanding multiple possibilities and their own preferences.
From a capital market perspective, Decitron Decision Engine further strengthens Zhongke Wenge’s differentiated tag of “decision intelligence.”
Currently, competition among AI companies is shifting from model parameters and general capability towards entering real business scenarios, solving complex problems, and achieving sustainable delivery. Compared to simply pursuing conversational ability, Zhongke Wenge has chosen a more enterprise-oriented, scenario-driven, decision-focused path.
For Zhongke Wenge, Decitron Decision Engine is both a product launch and an important window to showcase its long-term technical roadmap at the IPO milestone. As AI moves from generative to decision-making applications, the focus of enterprise AI competition may change from “who can answer better” to “who can help clients make better judgments.”
Deeper into the Industry
For Zhongke Wenge, entering the capital market doesn’t mark the end of its startup phase, but the start of a different long-term competition.
Behind the impressive growth curve lies an unavoidable financial reality.
From 2023 to 2025, Zhongke Wenge’s revenue was 250 million, 318 million, and 405 million yuan respectively, with a compound annual growth rate of 27.4%, but total losses over three years reached 583 million yuan.
The core reason for current losses is R&D investment; nearly 500 million yuan spent on R&D over the three years.
However, one data point is worth special attention. Zhongke Wenge’s average AI service delivery cycle shortened from 185 days in 2023 to 80 days in 2025.
The shortening of the delivery cycle means the reuse rate of underlying modules increased greatly, proving Zhongke Wenge’s ability to migrate systems and potentially forming an important competitive barrier in future.
But the enterprise AI market is fiercely competitive, “decision intelligence” as a concept still needs time to gain traction, and the progress towards platformization needs more financial indicators for continued validation.
Wang Lei is clearly aware of these issues. In conversation, he references Palantir—a company also founded on complex organizational data governance and decision support, recently attracting much capital market attention.
Meanwhile, Zhongke Wenge also keeps an eye on developments at global advanced AI companies like Anthropic. For the company, benchmarking does not mean simple copying, but finding its own enterprise AI implementation path based on China’s market, client structure, and industrial digitalization process.
Wang Lei said, if one day they truly succeed with “decision large models,” he hopes Zhongke Wenge will “leave a piece of AI memory for this era.”
The release of Decitron Decision Engine is an important step toward this ideal. For the newly listed Zhongke Wenge, “general decision” is no longer just a technical direction, but is becoming the core product narrative for the future.
Risk Notice and DisclaimerThe market is risky; investments must be made cautiously. This article does not constitute personal investment advice and does not take into account individual users’ specific investment objectives, financial situation, or needs. Users should consider whether any opinions, views, or conclusions in this article fit their particular circumstances. Investment based on this is at your own risk.