On May 15th, TapTechNews reported that today, ByteDance officially launched the Doubean large model, announcing that the price of large models has entered the 'per mil era'.
Tan Dai, President of ByteDance's Volcano Engine, introduced that after a year of iteration and market validation, the Doubean large model has become one of the largest in terms of usage and the most widely used large models in China. Currently, it processes 120 billion tokens of text and generates 30 million images per day.
'Large usage can polish a good model and greatly reduce the unit cost of model inference. The pricing of Doubean's main model in the enterprise market is only 0.0008 yuan per thousand tokens, which can process more than 1,500 Chinese characters for 0.8 mil, 99.3% cheaper than the industry.'
TapTechNews found that the Doubean large model, originally named 'Yunque', is one of the first large models in China to pass algorithm registration. Currently, ByteDance has developed AI assistant 'Doubean', AI application development platform 'Kouzi', interactive entertainment applications 'Maoxiang', as well as AI creation tools like Xinghui and Mimeng based on the Doubean large model, integrating it into more than 50 businesses including TikTok, Tomato Novel, Feishu, and JuEngine.
Zhu Jun, Vice President of ByteDance's Products and Strategy, stated that while technology is evolving rapidly, the core needs of users have not changed, including efficient information acquisition, work efficiency improvement, self-expression, social entertainment, and more. ByteDance has done a lot of learning and exploration, hoping to find a suitable form of large model application that provides convenience to users' lives and work in a natural interactive manner.
ByteDance also emphasized that Doubean is a major investment in large model applications. The Doubean app is currently ranked first in the AIGC category applications in the Apple App Store and major Android app markets in terms of downloads. According to Zhu Jun, there are already over 8 million intelligent beings created on Doubean, with 26 million monthly active users.
While serving ByteDance internally, the Doubean large model is also collaborating with many enterprise customers of Volcano Engine. Tan Dai believes that enterprises need models with different characteristics in different business scenarios to achieve the best results and cost-effectiveness. Therefore, the Doubean large model offers a family of models with multimodal capabilities, including nine main models such as general model pro, general model lite, speech recognition model, speech synthesis model, and WenShengTu model.
Tan Dai believes that reducing costs is a key factor in driving large models quickly into the 'value creation stage'. The pricing of the Doubean large model is significantly lower than industry prices: for example, the Doubean general model pro-32k version has a model inference input price of only 0.0008 yuan per thousand tokens, while similar models on the market are generally priced at 0.12 yuan per thousand tokens, 150 times the price of Doubean models.
According to the pricing announced by the Volcano Engine, one yuan can buy 1.25 million tokens of Doubean's main model, equivalent to about 2 million Chinese characters, or roughly three books of 'Romance of the Three Kingdoms'.
Tan Dai stated that lowering costs cannot only provide a lightweight version at a low price, but the main model and the most advanced models should also be cheap enough to truly meet the complex business scenario needs of enterprises, fully validate the application value of large models, and catalyze innovation beyond existing products and organizational models.
In addition, ByteDance's Volcano Engine also announced a collaboration with the China Electric Vehicle One Hundred People's Association to jointly establish an Automotive Large Model Ecology Alliance with more than 20 manufacturers including Geely, Great Wall Motors, Jetour, Sailes, Zhiji Automobile, etc.
At the same time, the Volcano Engine has formed the Intelligent Terminal Large Model Alliance with OPPO, Vivo, Honor, Xiaomi, Samsung, and Asus. Applications such as OPPO's Xiao Bu Assistant, Honor's Smart Office Intelligent Assistant, Xiaomi's Xiaoi Tongxue, and Asus's Douding AI Assistant for laptops have all integrated the large model services of the Volcano Engine.
ByteDance stated that the 'Kouzi Professional Edition' is currently integrated on the Volcano Engine's large model service platform 'Volcano Ark', providing enterprise-level SLA and advanced features, including enterprises like China Merchants Bank, Haidilao Hotpot, Super Orangutan, Liepin, as well as prestigious universities like Fudan University and Zhejiang University.