OpenAI announced that it will carry out a “limited-time promotion” for the cutting-edge model GPT-5.6 Sol, and the price of some services will be reduced by more than 20%. This is the second round of OpenAI's sharp price cuts in nearly a month — at the end of July, the GPT-5.6 Luna price of the same series had just been cut 80%. OpenAI's official explanation for this is that technical optimization has reduced inference costs. The Financial Times's analysis is “sobering” — customers are turning to open source big models provided by Chinese companies, and big US model companies “have to” cut prices. For some time now, people have been used to linking open source to low prices. Some researchers have discovered a rule — every step forward in the open source model puts pressure on closed source vendors to reprice. Researchers also regard this law as part of “Moore's Law” in the field of artificial intelligence. The only thing that gave rise to this law was an open source release in China. This time, OpenAI announced a price reduction, and it seems that there is also a reason for this. But the reason for this seems to be more than just the price—the open source model also provides more “value.” Chen Tianhao, a senior associate professor at Tsinghua University who participated in the “second-track” AI dialogue and exchange between China and the US, told Lord Tan that for cutting-edge business models, there are very few cases where training data is fully disclosed. In this reality, in addition to open model weights, it is also important whether to disclose data processing methods, training recipes, and engineering implementations. When Chinese companies released the open weighting model, they also revealed quite a few key training methods, engineering techniques, and infrastructure innovations through technical reports and open source code. This is not just the opening of model products, but extends the openness further to capability formation mechanisms and engineering tool chains. The deep meaning of this openness is that it has lowered the threshold for exploring advanced artificial intelligence engineering capabilities.

OpenAI announced that it will carry out a “limited-time promotion” for the cutting-edge model GPT-5.6 Sol, and the price of some services will be reduced by more than 20%. This is the second round of OpenAI's sharp price cuts in nearly a month — at the end of July, the GPT-5.6 Luna price of the same series had just been cut 80%. OpenAI's official explanation for this is that technical optimization has reduced inference costs. The Financial Times's analysis is “sobering” — customers are turning to open source big models provided by Chinese companies, and big US model companies “have to” cut prices. For some time now, people have been used to linking open source to low prices. Some researchers have discovered a rule — every step forward in the open source model puts pressure on closed source vendors to reprice. Researchers also regard this law as part of “Moore's Law” in the field of artificial intelligence. The only thing that gave rise to this law was an open source release in China. This time, OpenAI announced a price reduction, and it seems that there is also a reason for this. But the reason for this seems to be more than just the price—the open source model also provides more “value.” Chen Tianhao, a senior associate professor at Tsinghua University who participated in the “second-track” AI dialogue and exchange between China and the US, told Lord Tan that for cutting-edge business models, there are very few cases where training data is fully disclosed. In this reality, in addition to open model weights, it is also important whether to disclose data processing methods, training recipes, and engineering implementations. When Chinese companies released the open weighting model, they also revealed quite a few key training methods, engineering techniques, and infrastructure innovations through technical reports and open source code. This is not just the opening of model products, but extends the openness further to capability formation mechanisms and engineering tool chains. The deep meaning of this openness is that it has lowered the threshold for exploring advanced artificial intelligence engineering capabilities.