AI Wars | Mainland AI big models cut prices to attract customers, Alibaba cuts 97%, Baidu simply offers free services, ByteDance welcomes joint efforts to reduce costs

[ad_1]

Contents

Hope to guide explosive applications in various industries Byte’s attack took less than a week to provoke counterattack Support EJ Tech

The mainland AI giants launched a price war. Less than a week after ByteDance entered the market with a low price, Alibaba (09988)’s Alibaba Cloud counterattacked yesterday, announcing significant price cuts for nine commercial and open source models under its “Tongyi Qianwen” model. Among them, the price of the main model benchmarking OpenAI GPT-4 was reduced by 97%. A few hours later, Baidu (09888)’s Baidu Smart Cloud followed up and announced that its two main models of Wenxin Model, ENIRE Speed and ENIRE Lite, are all free, effective immediately.

Hope to guide explosive applications in various industries

Alibaba Cloud stated that it hopes to accelerate the explosion of AI applications through this significant price reduction. It predicts that the average daily number of large model application programming interface (API) calls this year is expected to exceed 10 billion times.

Alibaba Cloud announced price cuts at the Wuhan AI Leaders Summit yesterday. The overall price of the seven commercial and open source models of “Tongyi Qianwen” ranged from 25% to 97%, and two other open source models were opened for free use for a limited time of 7 days. Among them, the input price of Qwen-Long, the main model of GPT-4, dropped sharply from 0.02 yuan (RMB, the same below) per 1,000 tokens (the basic unit of language text, 1 Token is approximately 1 Chinese character) to only 0.0005 yuan/thousand tokens, a drop of 97%; the output price also dropped sharply by 90%, from 0.02 yuan to 0.002 yuan/thousand tokens.

Alibaba Cloud explained that this means that users pay 1 yuan to buy 2 million tokens, which is equivalent to the text volume of 5 “Xinhua Dictionaries”; while the input price of GPT-4 is about 0.22 yuan/thousand tokens, which is Alibaba Cloud hundreds of times.

Liu Weiguang, President of the Public Cloud Division of Alibaba Cloud Intelligence Group, believes that the cost of AI inference must drop 10 times or even 100 times every year to promote the explosion of AI applications in various industries. He said that this price reduction is due to the scale effect of Alibaba Cloud’s public cloud, and the reason why the price of output is higher than that of input is because output requires more computing power and the commercial value of the output content is higher.

In response to Alibaba Cloud’s move, Baidu Smart Cloud announced yesterday that Wenxin Model’s two main models, ENIRE Speed and ENIRE Lite, will be free of charge. It is reported that these two models are relatively lightweight large models released by the group in March this year, and are suitable for customers to fine-tune low-cost models for specific scenarios.

The market generally believes that Alibaba Cloud’s large price cut is in response to ByteDance’s earlier launch of services at a low price. Last Wednesday (15th), Volcano Engine, a cloud platform owned by ByteDance, announced that its self-developed bean bag large model is officially available for external services, and its pricing is significantly lower than that of its peers. The input price of its general model pro-32k is 0.0008 yuan/ Thousands of tokens, 99.3% cheaper than the industry average. After Alibaba Cloud’s price reduction this time, the price has been lower than that of Doubao.

Byte’s attack took less than a week to provoke counterattack

The head of ByteDance’s Volcano Engine responded that they very much welcome the price reduction of the “Tongyi Qianwen” large model, and will jointly help companies explore AI transformation at a lower cost and accelerate the implementation of large model application scenarios.

The industry believes that as the performance of large models improves, AI application innovation is entering a period of intensive exploration. However, high costs are still the key to restricting the large-scale application of large models. Therefore, the low-price strategy adopted by many major manufacturers this time will help By lowering the threshold for enterprises to use AI services and accelerating the popularization and application of AI technology, it is believed that more manufacturers will be forced to join price reduction wars to maintain competitiveness.