ModelBest has started the mandatory tutoring process for a potential mainland China initial public offering, marking a new financing step for a company built around smaller AI models designed to run locally. The Beijing startup has hired Citic Securities for the process, which must run for at least three months before a formal listing application can be filed. The South China Morning Post reported that ModelBest has not disclosed the proposed exchange, timing, fundraising target, or other transaction terms. The news is therefore not an IPO filing, much less a completed listing. It is an early but formal sign that the company is preparing for public markets.
ModelBest’s potential listing is notable because its strategy differs from the most visible part of the AI boom. Instead of centering its identity on the largest possible cloud model, the company has focused on compact systems that can run on local devices, including phones, laptops, and vehicles. That approach is often described as edge AI. It prioritizes low latency, privacy, reduced network dependence, and the ability to work on available hardware. In China, it also carries a strategic dimension: smaller, more efficient models can be optimized for domestic processors and may reduce exposure to shortages of the most advanced US chips.
The company was founded in August 2022, before the current generation of Chinese AI labs became household names. Its MiniCPM line has made it a recognizable player in the small-model field. Pandaily reported that MiniCPM had exceeded 38 million downloads across GitHub and Hugging Face, while a mid-July funding round put the company’s valuation above 20 billion yuan. Those figures are reported by Pandaily, not disclosed in the SCMP IPO report, and should be viewed in that context. What the mainland tutoring process confirms is that ModelBest now sees a public listing as a possible route to fund its next stage.
The prospect adds another dimension to China’s AI capital market. EastFrontier has covered how Chinese AI startups are preparing for public markets. ModelBest’s case shows that investors are not only looking at giant foundation-model developers. They are also looking at companies that can make AI usable on ordinary devices and within the hardware constraints of China’s technology ecosystem.
IPO tutoring is a preliminary but consequential milestone
Mainland IPO tutoring is a regulatory preparation process in which a prospective issuer works with a securities firm on governance, controls, disclosure, and compliance before applying to list. It does not guarantee a filing, approval, or completed listing.
For an AI startup, that preparation can be particularly demanding. Public-market investors will want to understand revenue quality, customer concentration, computing costs, intellectual property, data practices, model performance, competition, and the durability of any technology advantage. They will also want to know whether the company’s small-model strategy can create a business large enough to justify public-market expectations.
ModelBest has not said where it intends to list. China’s mainland boards offer different routes for technology companies, while Hong Kong has also become an important venue for AI and semiconductor listings. The lack of disclosed terms is normal at this stage, but it means any estimate of fundraising size or market capitalization would be speculative.
The decision to begin tutoring nevertheless suggests confidence that the company can articulate a credible commercial story. Investors will not only evaluate the quality of MiniCPM models. They will assess whether edge AI can become a repeatable source of revenue through enterprise deployments, device partnerships, software services, automotive applications, or other distribution channels.
Small models fit China’s hardware and deployment realities
The appeal of compact models is straightforward. Large cloud models can be extremely capable, but they often require substantial computing resources, network connections, and ongoing inference spending. A smaller model that runs on a phone, laptop, car, or local server can offer faster response times and keep some data closer to the user. That can be valuable for privacy-sensitive tasks, intermittent connectivity, real-time functions, and organizations that cannot send every request to a public cloud.
ModelBest’s strategy also fits the Chinese hardware environment. Access to the highest-end AI accelerators remains constrained by US export controls. Chinese companies are investing heavily in domestic chips and data centers, but compute remains expensive and scarce. Efficient models can reduce the amount of hardware required for a given task, making them attractive to device makers and enterprises that need AI features without operating giant model clusters.
This does not mean compact models will replace frontier models. Different workloads require different systems. A sophisticated research task, an extensive coding project, or a complex multimodal workflow may benefit from a large cloud model. A local assistant, device-control function, automotive feature, or offline tool may be better served by a smaller model. The market is likely to include both, with companies competing over where the boundary lies.
For ModelBest, the key challenge is converting technical efficiency into a defensible business. Open-source competition is intense. Alibaba, DeepSeek, Zhipu, Tencent, ByteDance, and global labs all release models that can be adapted for local use. A smaller-model specialist must demonstrate superior performance per unit of compute, reliable tools for deployment, meaningful customer partnerships, and a clear reason to choose its models over free or bundled alternatives.
Public investors will test the edge-AI thesis
A potential listing would give ModelBest access to capital, visibility, and a public valuation benchmark. It would also expose the company to a different standard of scrutiny. Private investors can support a long technical runway in the hope of future scale. Public investors will ask for clearer evidence of revenue, margins, customer adoption, and risk management.
China’s AI sector is entering a period when startups are trying to move from research credibility to corporate maturity. ModelBest’s case will depend on whether it can show that its models are embedded in devices and enterprise products where customers see immediate value.
Edge AI also provides a practical narrative for investors concerned about infrastructure costs. Running more capability locally can reduce cloud inference spending, though it shifts requirements to device processors, memory, software optimization, and support. The economics will depend on the specific deployment. But the core proposition is attractive: make AI useful without requiring every interaction to travel to a large, expensive data center.
ModelBest’s tutoring process shows that the company believes it is ready to make the case to a wider set of investors. Its potential IPO will test more than one startup’s valuation. It will test whether China’s market rewards AI strategies built on efficiency and local deployment.
For now, the company remains at a preparatory stage. No filing date, exchange, or fundraising target has been disclosed. Yet the move is a meaningful marker in China’s AI financing cycle. The industry is not only producing larger models and bigger data centers. It is also producing companies that argue smaller, smarter, locally deployable models may be one of the most commercially durable forms of AI.