tongyi qianwen reduces prices again zhou jingren: prices are not too low, but still too expensive for future applications
2024-09-20
한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina
the 2024 yunqi conference was held in hangzhou from september 19 to 21. during this period, alibaba cloud cto zhou jingren accepted interviews from media including beijing news shell finance.
at this conference, the tongyi qianwen large model ushered in another update. in addition, after the first substantial price reduction in may, the three tongyi qianwen main models on the alibaba cloud bailian platform were reduced in price again. for example, the price of qwen-turbo dropped by 85% to as low as 0.3 yuan per million tokens. regarding whether the price reduction of the model may be too low, or even trigger a price war, zhou jingren said that there is no such thing as "low enough" price, and alibaba cloud has fully invested in upgrading ai infrastructure. talking about the o1 model released by openai, he thinks it is admirable and also points out a path for the development of large models.
zhou jingren was interviewed by beijing news shell financial reporter luo yidan/photo
price reduction is a "very serious process" and is consistent with the model open source strategy
"we reduce prices mainly through technical optimization, including cloud infrastructure improvement and model structure improvement, which can further reduce the model's inference cost. only in this way can we more effectively develop various ai applications." zhou jingren said in an interview, "every time we reduce prices, it is a very serious process. we have to weigh various aspects such as the development of the entire industry, feedback from developers and corporate users, etc. this is consistent with our model open source strategy."
shell finance reporter learned that since the first announcement of the price cut, the number of paying customers of alibaba cloud bailian has increased by more than 200% compared with the previous quarter. a number of enterprises and developers have given up private deployment and chose to directly call various ai models on bailian. however, including alibaba cloud, the competitive price cuts and even free strategies of various cloud vendors also ushered in a wave of "price wars" in the industry in may. will the price cut again this time trigger another "price war"?
in response, zhou jingren said that he did not think this was a price war. he cited the example of mobile internet charges, saying, "how much does it cost to surf the internet today? my mobile internet package is less than 200 yuan. today, 200 yuan can use dozens of gb of traffic. if i use this kind of traffic in 20 years, i will probably go bankrupt. therefore, today's price is not low enough. compared with the huge applications in the future, it is still too expensive."
zhang qi, vice president of alibaba cloud, added that today's ai is equivalent to the internet around 1996, that is, in the bbs era. at that time, internet access was very expensive. later, the internet and mobile internet continued to develop, and operators vigorously built infrastructure. now we are also building ai infrastructure. only by lowering the fees can we talk about the explosion of applications in the future. "so this long-term goal is what we consider more, rather than having to calculate today, how much money we can make immediately, or what goal we will reach immediately."
alibaba cloud is fully committed to upgrading ai infrastructure and aims to become an "infrastructure provider" in the ai era
not long ago, openai launched an upgrade of the o1 model that focuses on enhancing the reasoning ability of large models. this time, tongyi qianwen also upgraded its model. for the technical solutions called "system 2" such as "chain of ideas" proposed in the technical documentation of openai's o1, zhou jingren said that this is a very representative and revolutionary direction. "this series of capabilities can make it so robust. i think it is still very admirable and it also points out a feasible path."
zhou jingren said that alibaba cloud is setting a new standard for ai infrastructure in the ai era, comprehensively upgrading the technical architecture system from servers to computing, storage, networking, data processing, model training and reasoning platforms, making the data center a supercomputer and providing high-performance and efficient computing services for each ai and application.
at the conference, he demonstrated the ai-driven upgrades of alibaba cloud's entire product family, including the latest panjiu ai server, the gpu container computing power launched for the first time by alibaba cloud acs, alibaba cloud cpfs file storage, and the artificial intelligence platform pai.
shell finance reporter noticed that there were many humanoid robot companies invited to the yunqi conference booth. at present, the embodied intelligence of large models + humanoid robots is a popular concept. so does alibaba cloud have any considerations for embodied intelligence? zhang qi said, "today, many robot startups are using tongyi's open source or closed source models, installing them on robots to become the facial features and brain of robots. this is our relationship with the industry. it doesn't mean that alibaba cloud has to make autonomous driving and robots by itself. for example, xiaopeng's p7+ was unveiled at the yunqi conference this time. all its computing power solutions for autonomous driving are in cooperation with alibaba cloud. we provide xiaopeng with computing power and cockpit model capabilities. this is alibaba cloud's position today."
"alibaba cloud is ready to meet the exponential growth in demand for gpu computing power, especially the upcoming explosion in the inference market," said zhou jingren.
luo yidan, financial reporter of beijing news shell
editor: yue caizhou
proofread by mu xiangtong
