With the global success of ChatGPT, the domestic large model industry has followed suit with a series of competitions involving parameters, resources, capital, and long-text processing capabilities. This competition has turned into an intense price war.

Industry insiders have expressed diverse views on the sudden wave of price reductions for large models. Some believe that pricing strategies depend on the specific business model of the company. If the service is aimed at businesses, the price cut may not be for the direct sale of the model, but rather to promote cloud service sales and help businesses transition from traditional services to new platforms. But for other companies, facing stiff competition from giants, a defensive strategy may be necessary for the time being.

Regarding the price war, some view it as a marketing tactic. In fact, for the current large model enterprises, the revenue from APIs does not constitute a major part, so price reductions will not have a significant impact on their revenue.

The outbreak of this wave of price reductions is behind a war over ecosystem dominance. Although changing large model providers seems to involve only a simple change of URL, there are significant differences in performance and focus between different models. Once a business has adapted to a particular model, there may be little incentive to switch to another, especially when there is an actual cost to switching. Therefore, many companies are willing to attract users through price reductions to build up a user base and ecosystem.

Industry experts also pointed out that the reduction in prices of large models is a trend in the whole industry. Taking GPT-4 as an example, its launch price was half that of GPT-Turbo, providing valuable lessons for the domestic large model industry. By lowering price barriers, large model manufacturers hope to attract more enterprise users and individual developers to use their technology. This not only helps to balance income and costs but also accelerates the explosive growth of AI applications.

In the current market environment, early planners have the opportunity to win a higher market share. Lowering prices at this time can quickly accumulate a user base, gather an ecosystem, and consolidate their status in the industry.

Industry experts Qi Tongquan and Wang Xiaochuan hold different views on the prospects of artificial intelligence. Wang Xiaochuan expressed that although people are very hopeful about AI's potential and do not want to miss any opportunities, he still advises startup companies to be cautious, suggesting that the high cost of large models and price declines will not change the production relationship based on supply and demand networks. Some industry insiders see the current price reduction in large models as a redux of cloud service providers fighting for the market through cash burn in years past. Wang believes that although the development of cloud services in China is not fully satisfactory, the rise of large model technology has provided new productivity for cloud services. In the future, cloud service providers are expected to move away from price competition, but when a clear industry change will occur remains unknown.

On the other hand, Qi Tongquan mentioned that the reason the domestic cloud computing industry is engaging in a price war is due to the slowdown in internet industry development, slowing user growth, and product and service homogenization making price a key competitive factor. By contrast, the large model industry is still in the early stages of development and far from reaching the peak of growth. QuantumBit Think Tank projects that China's generative AI market could reach 1.1491 trillion yuan by 2030, with an expected compound growth rate of 42%, and the market size could reach 1.3 trillion USD by 2032. The current price reductions are more a strategy by manufacturers to seize the opportunity, attract users and expand scale.

Industry leader Li Kaifu emphasizes that the large model industry must not repeat old mistakes and embark on a wasteful path of capital burning. He predicts that the annual cost of large models will be significantly reduced, but the call ratio of industry service models is still low. An anonymous chief technology officer at a large model start-up company indicated that the current price reductions are aimed at attracting enterprise-level developers. These companies need developers to build ecosystems, feedback their training results, while the developer community needs low-cost large model technology to ease their financial burdens.

Despite open-source large models providing a completely free choice, they come with deployment and application barriers, such as the need for professional engineering capabilities in the areas of computing power, model deployment, and optimization, which are not available in all companies. However, large model enterprises can provide APIs to help clients use and solve specific problems directly, for example, cloud service providers can offer mature service capabilities like tool optimization platforms. Similarly, C-end developers can use these technologies to develop innovative applications and distribute them through platforms with traffic advantages, such as ByteDance's Volcano Engine.

However, such work is not an easy task, there is no absolutely free lunch in the market, nor is there a completely free large model technology. Some developers complain that although the price of domestic large model APIs has dropped or even offered for free use, there are usually multiple layers of hidden logic in actual operation. Typically, it's a bait-and-switch where the high-end services are initially free, and then you are forced to use other fee-based projects.

In today's business competition, the entry barrier for the B-end market is gradually lowering, especially in the field of technology applications. Some manufacturers offer free or more affordable versions to quickly popularize their products. However, these versions usually support a smaller scale of parameters and computing power, and are more suitable for less frequent training tasks or small-scale reasoning tasks.

Facing the advent of the large model era, how enterprises choose their business strategies has become the focus of industry discussion. Experts believe that in the closed-source model market, if the price of a product is lower than that of open-source models, it may be because its scale is relatively smaller. In the Chinese business environment, the commercial route of selling single products through API services is not favored because compared to the C-end market, China's ToB market is not large, and ToB transactions often involve RMB, while costs are calculated in US dollars.

In this context, the industry's expectations for super applications have not weakened. Although the discounting of large model APIs may only be the initial phase of market adjustment, there is still room for price increases in the application end in the future. Industry professionals widely agree that the current downward price trend is beneficial to the development of the whole industry.

In terms of strategy choice, large model manufacturers aim to develop applications that can satisfy To small B businesses and also be applicable to the C-end market. Against such a backdrop, pricing strategy becomes key to attracting developers and enterprise customers. Capital market experts believe that free large models can quickly drive traffic, but in the long run, more personalized services are needed to attract paying corporate users.

In the AI industry, risks and opportunities coexist. For businesses, reducing costs and improving the quality of applications are the current focal points. The president of Volcano Engine pointed out that due to the significant enhancement of technical capabilities, it has become crucial to develop suitable applications. The CEO of Youzan analyzed the logic behind application-side price increases and pointed out that if the SaaS industry tends towards stability, then why is there a need to decrease prices? Enterprises that can truly stand in the model application market must possess unique business scenarios and delivery capabilities. There are not many such players, hence, there is no motivation to reduce prices.

Ultimately, companies that treat AI technology as a productivity tool, if they wish to develop over the long term, must introduce truly valuable products and formulate corresponding charging standards in line with market demand. A senior executive from Youzan stated that even though the cost of productivity tools continues to decrease, the price of products with higher technical content is constantly rising, therefore the value of AI technology on the application side will become increasingly prominent.

In the view of industry expert Wang Xiaochuan, while low cost is a significant advantage, this point is undoubted. However, if this price advantage is to be transformed into a core competitive strength in the market, this alone is far from enough. He believes that for startups, the key lies in building a "super model" and relying on a "super application" to realize a strategy of dual drive, and not simply depending on application programming interfaces (APIs) to solve all problems.

Wang Xiaochuan further pointed out that if a "super application" wants to stand in the era of big data, its application scale needs to achieve a qualitative leap—the daily active user count should grow from 1 million to 100 million, with even a 3.3-fold room for fluctuation on this basis. This means that the daily active user count of a truly competitive super application should be between thirty million and three hundred million. Only such scale figures are worthy of the standard for super applications in the "big model era."