GuShiio.com 
Tan Dai, President of Volcano Engine
Tantai, President of Volcano Engine, responded for the first timeBaiduShen Dian ‘s views on the low price of large models and the impact of DeepSeek.
GuShiio.comAGI reported on February 13 thatShen Dian, executive vice president of Baidu Group and president of Baidu’s intelligent cloud business group, said at the all-staff meeting that the malicious price war of domestic large models last year caused the industry’s overall revenue generation to be several orders of magnitude worse than that of foreign countries.
Shen Dian made it clear that in the face of DeepSeek’s onslaught, the first AI product to bear the brunt is the bean bag with ByteDance, because its training costs and streaming costs are high.
In this regard, GuShiio.comAGI learned that Tan Dai, president of Volcano Engine, responded through a circle of friends that the price reduction of large models is achieved through technological innovation, and everyone should focus on basic skills like DeepSeek and avoid making groundless guesses.
Tan Dai said that the pre-training cost and reasoning cost of the Doubao 1.5 Pro model are lower than those of DeepSeek V3, and far lower than other domestic models. It has a very good gross profit at the current price.
“Domestic and foreign manufacturers are relying on technological innovation to reduce model prices. We have only achieved the price level of Gemini 2.0 Flash. This price is entirely based on technological progress.& quo; Tan needs to further appeal,“Everyone should focus on basic skills and innovation like DeepSeek, be not impatient or impatient, less groundless speculation, and attribute blame to outsiders.” rdquo;
It is reported that the Doubao-1.5-pro-32k’s million-token input price is 0.8 yuan, the hit cache is 0.16 yuan, and the output price is 2 yuan. It was disclosed that its gross profit margin reached more than 50%.
Data shows that the number of calls for the bean bag model has increased rapidly in the past year. The average daily usage of tokens in December last year exceeded 4 trillion yuan, a 33-fold increase from when it was first released in May.
It is worth noting that the volcanic engine recently announced that it has been connected to DeepSeek’s model. At the same time, on February 12, the Doubao Big Model team also announced a new sparse model architecture UltraMem, which is 2-6 times faster than the MoE (Mixed Expert Model) architecture, and can reduce the reasoning cost by up to 83%. It continues to exert efforts in technological innovation.
