Financial AI models have never lacked acclaim.

Over the past year, nearly every well-known technology company has released a financial AI model. However, only a handful have secured institutional orders and successfully deployed them in production environments.

While the excitement is macroscopic, the differentiation is microscopic—Ant Group, Du Xiaoman, Alibaba Cloud, Huawei Cloud, and Tencent have all moved in different directions.

Ant Group's Bailing Model: Leveraging Native Scenarios, with Intelligent Agents as a Game-Changer

Ant Group's AntFinGLM (Bailing Model) starts with a natural advantage over others.

The barriers to financial native scenarios are extremely high. With Alipay, the Bailing Model has access to transactional, credit, and small business operational data. Hundreds of millions of daily real transactions serve as its best training material.

In terms of specific functions, the Bailing 2.0 Model of MYbank has been applied to the daily operations of tens of millions of small businesses—high-frequency functions such as credit approval, bill processing, and financial and tax consulting are all packaged into a single entry point.

Rich B-end implementation cases. The promotion of Bailing products at the B-end is also accelerating. Ant Group has signed contracts with multiple joint-stock banks and city commercial banks. Ant Group's executives also stated at last year's Financial Street Forum that the company's main cooperation model is not charging per Token but based on business outcomes.

This model is more popular in the financial sector. Banks care less about how many times you adjust the model and more about whether risk control improves and efficiency increases. Charging based on usage effectiveness is also an expression of product confidence.

Outstanding professional capabilities of intelligent agents. Ant Group also deployed another AI piece this year. In September, it released the new-generation financial AI model Ling-3.0-flash-Fin. This product's main feature is not Q&A interaction but the ability to complete an entire set of investment research tasks from start to finish. This is an extension of Ant Group's capabilities beyond Bailing.

Complete security governance system. Besides the model itself, Ant Group has also addressed another deeper issue: Why do financial companies trust AI? Ant APASS intelligent agent was born for this purpose—not to generate content but to answer three questions: Is this AI reliable? Which client does it come from? What business is it allowed to conduct?

However, the advantages of the Bailing Model almost entirely rely on Ant Group's ecosystem. Can it still compete without Alipay and MYbank? The answer lies in two weaknesses.

Lack of independent profitability. In Ant Group's financial solutions, the Bailing Model is merely a functional component, and the total revenue from model calls is relatively low. Ant Group prefers to sell the entire solution, making it difficult for the Bailing Model to operate independently.

Shortcomings in deep investment research capabilities. The Bailing Model excels at general financial tasks but appears less professional in complex tasks such as in-depth understanding and valuation derivation. After all, Ant Group still lags behind established companies like Hundsun and Wind in the financial investment research scene.

Du Xiaoman's Xuanyuan Model: Focusing on the Credit Sector, the First Choice for Small and Medium-Sized Banks

While Ant Group's Bailing Model aims for comprehensive coverage, Du Xiaoman's Xuanyuan Model takes the opposite approach, focusing solely on the credit sector.

First-tier risk control capabilities. Du Xiaoman is one of the earlier Internet finance companies in China, with experience in every aspect of credit, from approval and anti-fraud to post-loan management. Its industry experience is relatively rich.

In essence, the Xuanyuan Model can be seen as Du Xiaoman's "lessons learned" report, which is vastly different from the "theoretical" risk control approaches of many companies crossover (cross-border) into financial AI.

Lightweight product structure. For small and medium-sized financial companies like city commercial banks and rural commercial banks, the Xuanyuan Model focuses on API calls and lightweight deployment, eliminating the need for customers to invest in expensive hardware upfront, accurately addressing the pain points of the lower-tier markets (lower-tier market).

Mature credit business functions. In the credit industry, paying tuition is inevitable. Therefore, the training method for the Xuanyuan Model is to first run through its own business before expanding outward, rather than using clients as test subjects. This approach may be slower but is less prone to errors.

Thanks to the improved capabilities of the Xuanyuan Model, Du Xiaoman has reduced the overall risk of credit delinquencies by about 50% in its internal AI credit review process, while the review time has been shortened from 10 minutes to just 30 seconds.

Early commercialization. The Xuanyuan Model does not rely on an overall solution and can be sold as a standalone product. Therefore, among several Internet giants, Du Xiaoman's vertically integrated financial model has the fastest commercialization speed, and the Xuanyuan Model has already secured many B-end paid contracts.

However, the single-sector advantage of the Xuanyuan Model is precisely its capability boundary.

Relatively single application scenario. The Xuanyuan Model performs well in credit risk control but lacks capabilities in other scenarios, such as investment research and corporate due diligence. Relying solely on a single sector also makes it difficult to sustain the long-term value of an independent AI model enterprise.

High early-stage cost investment. Although the Xuanyuan Model now generates revenue from orders, its high upfront R&D and operational sunk costs mean that the product has not yet achieved independent profitability.

Alibaba Cloud's Dianjin Model: All-Scenario Coverage, Supported by Cloud Ecosystem

Alibaba Cloud's Dianjin Model does not rely on a single business breakthrough but leverages a combination of cloud ecosystem capabilities.

Strong all-scenario coverage capabilities. The Dianjin Model focuses on the banking, securities, and insurance sectors, with independent working capabilities in various scenarios such as investment research, credit, and claims processing. It supports private deployment and MaaS calls, making it capable of meeting almost any demand.

Integrated advantages of the cloud platform. The Dianjin Model relies on Alibaba Cloud to form a complete functional system that integrates large-scale computing power, data governance, model training, and application deployment.

When customers purchase the model's capabilities, Alibaba Cloud can also provide other business functions. Moreover, most customers do not need to worry about computing power and deployment during the entire delivery process, making hassle-free service one of its biggest selling points.

Strong ecological compatibility. The Dianjin Model is highly open and can connect with third-party financial information databases, compatible with many internal systems of securities firms and fund businesses. It has also launched a payment financial AI model in cooperation with China UnionPay.

This creates a positive cycle: as its open ecosystem becomes richer, the scenario-based capabilities of the Dianjin Model will also improve.

Strong risk resistance. Alibaba Cloud's entire AI business has achieved continuous profitability. Although the Dianjin Model, as its vertically integrated financial AI product, has not yet independently generated profits, it is supported by Alibaba Cloud and does not need to reduce its scope due to short-term investment pressure, allowing it to gradually refine and patiently update.

However, the broader the scenario coverage of the Dianjin Model, the more difficult it is to maintain depth.

Insufficient deep risk control capabilities. The Dianjin Model excels in generalized office scenarios such as document preprocessing and content summarization but falls short compared to vertical players in tasks requiring "strong decision-making," such as credit risk control and anti-fraud.

After all, banks' core risk control systems will not be entrusted to a generalized model that "does a little of everything." To address this, the Dianjin Model still needs many targeted improvements.

Long project cycles. Alibaba Cloud's main customers are large financial companies that do not want a templated product but rather a system rebuilt from scratch. Such projects have long cycles, high investment, and high R&D requirements.

Huawei Cloud's Pangu Model: A Trusted Choice for Government and Enterprise Clients, with a Strong Focus on Domestic Innovation

Huawei Cloud's Pangu Financial AI Model takes a hardcore approach.

Fully domesticated stack, with clear advantages in domestic innovation. The Pangu Model achieves full-stack self-research from chips to frameworks, making it nearly unbeatable in meeting the "autonomous and controllable" demands of state-owned major banks and policy banks.

The tighter the geopolitical situation, the more valuable this advantage becomes. Currently, Huawei Cloud ranks first in the domestic financial AI model market share, with state-owned major banks and government and enterprise clients on its list.

Strong private deployment capabilities. The Pangu Model is among the first products to pass the financial AI model standard compliance verification, receiving the highest rating.

For public sector private deployments, Huawei is almost the default choice. In scenarios such as corporate due diligence and contract review, the Pangu Model already has mature implementation experience, with positive feedback from clients.

Outstanding large-project delivery capabilities. The Pangu Model offers comprehensive solutions for large financial institutions, with high single-project contract values and rich delivery experience. Half of China's top 20 major banks have already chosen Huawei's AI platform.

Rich industry template cases. The Pangu Model has accumulated thousands of financial industry scenario templates and incorporates extensive industry norms and knowledge bases, allowing clients to avoid building models from scratch. The actual effects are also intuitive (intuitive). According to Huawei Cloud's official disclosure, some bank clients have used it to simplify counter operations from five steps to one and improve credit report efficiency by more than half.

The flip side of pursuing a high-end route is the limitation of business scale.

Highly concentrated client base. The Pangu Model's main clients are concentrated among state-owned major banks and large institutions, with low penetration among small and medium-sized banks and securities firms, essentially abandoning the lower-tier market.

In the Pangu Model's financial joint innovation initiative, Guangfa Securities, Bank of Communications, and China Zheshang Bank were the first sponsors, all top-tier institutions. Public information reveals almost no cooperation cases between the Pangu Model and small and medium-sized financial institutions.

Insufficient ecological openness. The Pangu Model's ecological partners are mainly traditional financial IT service providers like Yusys Technologies, Nantian Information, and Advanced Digital Technology, focusing on information integration and delivery. Its compatibility with third-party toolchains and quantitative systems commonly used by quantitative hedge funds is generally limited.

Tencent WorkBuddy Financial Edition: Breaking Through with Office Scenarios and a Circuitous Approach

Tencent's WorkBuddy Financial Edition does not compete head-on with rivals but instead uses office scenarios and security capabilities as a fulcrum to gradually penetrate the financial industry.

Impressive performance in corporate credit scenarios. Under traditional due diligence models, client managers must jump between industrial, financial, and judicial systems, with report generation taking ten days to half a month. WorkBuddy Financial Edition streamlines these processes, reducing report generation from ten working days to one. It has already been implemented in hundreds of financial institutions, including CICC and Ping An Bank.

Deep accumulation of security and risk control capabilities. Tencent has invested heavily in anti-fraud over the years, with its AI anti-fraud assistant achieving an 89% success rate in social scenarios. Since financial black market activities and social fraud are often perpetrated by the same groups, the model capabilities accumulated in anti-fraud can be directly applied to financial risk control. This technological foundation is something pure financial vendors cannot catch up with in a short time.

Seamless integration with enterprise office ecosystems. WorkBuddy Financial Edition is deeply integrated with WeCom and OA systems, eliminating the need to switch between contract review and compliance inspections. AI is embedded in daily office workflows rather than existing as a separate application. Only by making it easy for frontline employees to use will they not resist it.

Strong multimodal processing capabilities. WorkBuddy Financial Edition can handle documents, images, voice, and video. It can recognize and extract key information from screenshots, handwritten documents, and even audio recordings sent by clients. In scenarios such as telephone customer service and video interviews, multimodal capabilities determine how much AI can replace human labor.

However, using office scenarios as an entry point ultimately only reaches the periphery.

No independent vertically integrated financial AI model. The majority of Tencent Cloud's revenue comes from general MaaS, with WorkBuddy Financial Edition being just one brand under it, making it difficult to calculate revenue independently. Moreover, WorkBuddy Financial Edition is essentially a general model with a financial industry shell, not specifically trained for the financial sector.

Lack of native financial data. WorkBuddy Financial Edition does not have access to real business data such as credit and transactions, so its deep reasoning capabilities naturally fall short compared to Ant Group and Du Xiaoman. It mostly assists outside core systems and cannot handle banks' internal money management and risk control core functions.

Financial AI Models: Who Will Go Further?

After examining the cards held by the five vendors, it's time to discuss the rules of the game itself.

Pure model capabilities are no longer a moat; business closure is the key to victory. The era of comparing parameters and benchmark scores is over. Financial institutions will not pay for a model that only answers questions. Only those that can be embedded in the full process of risk control, due diligence, and investment research will make it onto procurement lists.

Commercialization paths will clearly diverge, with only two types of players surviving: those backed by native financial businesses, like Ant Group and Du Xiaoman, and those relying on cloud platforms and domestic innovation capabilities, like Alibaba Cloud and Huawei Cloud. Pure model vendors without either advantage will see their space quickly squeezed.

Profitability in financial AI models does not come from "selling model calls" but from selling complete solutions. Currently, no major vendor's financial AI model is independently profitable. In the future, such products will not generate revenue through token-based charging but through packaged delivery of models, knowledge bases, deployment, and maintenance.

Domestic innovation and security compliance are hard requirements; pseudo-concept products will be eliminated. The financial industry has strict requirements for data security, model interpretability, and audit traceability. Products without compliance filings and private deployment capabilities will likely fail to access banks' core businesses.

Conclusion

Each of the five vendors has its own trump card, but having a good card does not guarantee victory.

Ant Group has the deepest scenarios but lacks an independent monetization path. Du Xiaoman excels in credit but has limited scenarios. Alibaba Cloud covers all scenarios but lacks depth. Huawei Cloud dominates in domestic innovation but has a closed client base. Tencent impresses in office scenarios but lacks native data.

The elimination round for financial AI models is not about who has the stronger model but who is closer to the business. Products that cannot access core business processes will ultimately remain just a PPT presentation.