[Global Times reporter Zhao Jue] The endless emergence of large models of domestic artificial intelligence (AI) is often described by insiders as a "100 model war". Now, the fierce competition in this industry has entered the stage of price war. Recently, the big model products of domestic cloud manufacturers have faced a storm of price cuts, including Alibaba Cloud, Baidu, the volcanic engine of Byte Beat, and Zhipu AI, which have announced price cuts, and some products are said to have dropped by as much as 99%. A number of industry experts told the Global Times on the 22nd that manufacturers hope to attract customers and seize the market by reducing prices, while reducing the use cost to promote the rapid landing and ecological prosperity of large model applications.
At the China International Nuclear Power Industry Exhibition, a booth demonstrated AI writing application. （ Visual China ）
The era of "pricing by per cent"
On the 22nd, Tencent Cloud announced that it will comprehensively reduce the price of large models, and its hybrid lite will be free from now on. Baidu has just launched a free measure for its big model on the 21st, causing hot debate.
Baidu announced that ENIRE Speed and ENIRE Lite, the two main models of Wenxin Big Model, will be free immediately, opening the "free era" of API (application programming interface) calls of the big model. According to the person in charge of Baidu, these two models support 128k and 8k context lengths, and are currently the models serving the most users in Baidu Wenxin's big model series. Alibaba Cloud and Baidu also significantly reduced the prices of 9 commercial and open source models of Tongyi Qianwen on the same day. Among them, the API of Qwen Long, the main model of benchmarking GPT-4, dropped from 0.02 yuan/thousand tokens (the smallest unit of text processing) to 0.0005 yuan/thousand tokens, which means that users can buy 2 million tokens for 1 yuan, which is equivalent to the text volume of five Xinhua Dictionaries. According to statistics, this price is about 1/400 of the GPT-4 price.
Before the above-mentioned big manufacturers announced the price reduction of big model products, Byte Beat held a press conference on the 15th to release a series of latest products such as the big model family of Byte Beat, Volcano Ark 2.0, AI applications and AI cloud infrastructure. According to the press conference, the big model of bean curd was officially opened for external services. The reasoning input prices of the general model of bean curd, pro-32k and pro-128k, in the enterprise market, decreased by 99.3% and 95.8% respectively compared with the industry. In addition, enterprises such as Wisdom Spectrum AI and Magic Square Quantification have also announced price reduction of their big models.
Some media described that with the successive price reductions of various manufacturers, the big model is moving from the pricing model of "pricing by points" to the era of "pricing by per cent".
Zhu Keli, executive director of the China Information Association, told the Global Times on the 22nd that the big models provided by cloud service providers are powerful engines for incubating AI applications. These big models are like smart brains, carrying mass data (Rights protection) And complex algorithms, providing a solid foundation for AI applications. Through the large model provided by cloud service providers, developers can develop AI applications more efficiently, reduce development costs, and shorten development cycles, thus promoting rapid iteration and widespread landing of AI applications.
In this regard, Tan Dai, president of the cloud service platform Volcano Engine under Byte Beat, said that the technology of the big model has significantly improved this year, and the application link is crucial. Ecological prosperity is the key to promotion and application. In the market, many customers are eager to try large models, but the innovation risk is high, especially AI risk. "To reduce uncertainty, we must significantly reduce the cost of trial and error, so that large models can be widely used, whether by large enterprises or individual developers. The market calls on us to provide core technologies with lower costs," Tan Dai said.
AI application popularization
Since the end of 2022, the wave of large models has experienced rapid development in the past year. According to the data disclosed by the National Data Bureau in March this year, the number of large models with a parameter scale of more than 1 billion in China has exceeded 100. Industry insiders believe that after a period of "frenzy", some manufacturers have begun to focus more on optimizing user experience, reducing use costs, and promoting the commercialization of large models to speed up the implementation, which is also an important background for the collective price reduction.
Some insiders, including Kai Fu Lee, believe that this year will be the first year of the explosion of large model applications. A report released by Tencent Research Institute this month believes that under the guidance of important policies such as "AI+", the industry model is expected to accelerate its application in traditional industries, and with the support of cloud intelligence integrated infrastructure, it will develop in the direction of multimodality, AI, end and miniaturization, and be more deeply embedded in the workflow of various industries, So as to promote the improvement of productivity.
Zhu Keli said that the general price reduction has brought great impetus to the implementation of large models and AI applications. Reduce the price to lower the threshold of AI application, so that more enterprises and individuals can access AI technology, thus promoting the popularization of AI applications. At the same time, price reduction also promotes market competition, stimulates the innovation vitality of cloud service providers, and constantly improves the performance and functions of large models to meet the needs of more users.
Hualong Securities Research News said that with the further fierce technical competition between large model manufacturers at home and abroad, the signal of opening price war in the large model industry is obvious. By lowering the price threshold, large model manufacturers are expected to attract a broader group of enterprise users, thus further balancing revenue and costs. At the same time, more consumer end users are expected to use basic AI applications for free. The huge number of visits will help enterprises further improve their model service capabilities and complete a virtuous circle.
Beyond the "price war"
Not only in China, but also in the past year, OpenAI, a leading enterprise in the field of large models in the United States, has cut prices for its products many times. In January this year, OpenAI announced that the input price of the GPT-3.5 Turbo model would be reduced by 50% and the output price by 25%. On the 13th of this month, OpenAI released a new generation of large model GPT-4o, which is said to double the speed and halve the price.
The AI technical director of a large Internet company in Beijing told the Global Times that the current price reduction of large models at home and abroad is an inevitable product of changes in market demand and competition pattern. After a period of concentrated research and development and "burning money", the market's demand for cost-effective AI products and services is increasingly urgent.
The above people believe that price reduction will intensify market competition and promote enterprises to continuously improve their technical level and service capabilities. In this process, enterprises that can provide more cost-effective services will stand out, while enterprises that cannot adapt to market changes may be out.
In addition to the price, strong ability and excellent experience are still important factors for the success of the big model product. Zhu Keli told the Global Times that although price war can capture market share to a certain extent, efforts should be made in many aspects to really accelerate the implementation of large model applications. For example, cloud service providers should continuously improve the performance and functions of large models to meet the diverse needs of users; We should strengthen cooperation with all walks of life, deeply understand the needs of the industry, and provide customized AI application solutions for different industries. It can not only improve the landing effect of AI applications, but also further expand the market share of cloud service providers.
"To really accelerate the application of the big model, cloud service providers need to make efforts in technology, cooperation, security, privacy protection and other aspects. Only in this way can the big model really serve all walks of life and promote social progress and high-quality economic development." Zhu said.
Massive information, accurate interpretation, all in Sina Finance APP
Editor in charge: Jiang Yuhan
