On July 24, Xiong Wei, a Chinese internet industry analyst at UBS Securities, said at a media conference that as demand is stratified, there are differences in pricing power between different echelon models. According to statistics, the training cost for China's leading models is about 1/10 of that of overseas leaders. In terms of inference costs, the Chinese model API price is about 10% to 20% of the average price of overseas benchmark models, and can still maintain a gross profit level of 20% to 40%, which is slightly lower than that of American peers, but it is still at a very healthy level. Moreover, this advantage does not stem from aggressive subsidies or crazy concessions from manufacturers, but rather structural cost efficiency advantages, including factors such as training algorithm innovation and architecture design, differences in model strategy priorities, and open source ecosystem collaboration.

On July 24, Xiong Wei, a Chinese internet industry analyst at UBS Securities, said at a media conference that as demand is stratified, there are differences in pricing power between different echelon models. According to statistics, the training cost for China's leading models is about 1/10 of that of overseas leaders. In terms of inference costs, the Chinese model API price is about 10% to 20% of the average price of overseas benchmark models, and can still maintain a gross profit level of 20% to 40%, which is slightly lower than that of American peers, but it is still at a very healthy level. Moreover, this advantage does not stem from aggressive subsidies or crazy concessions from manufacturers, but rather structural cost efficiency advantages, including factors such as training algorithm innovation and architecture design, differences in model strategy priorities, and open source ecosystem collaboration.