Major U.S. telecom operator AT&T stated that open models currently account for about 25% of the company's AI usage, and the company expects this proportion to rise to 70%–80% in the future. In certain application scenarios, switching from closed proprietary models to open models has already reduced costs by 80%–90%. Goldman Sachs noted that a clear model stratification is emerging on the enterprise side: tasks that do not require extreme reasoning capabilities can shift to lower-cost open models, while the most complex tasks will continue to rely on more powerful proprietary models. As domestic open models gradually become alternatives to U.S. models, the AI Token costs tracked by Goldman Sachs have dropped 49% since May. Model invocation is becoming increasingly affordable, which benefits enterprises by enabling them to expand AI usage, but foundational model providers will face greater pressure on unit pricing and profit margins. In the next phase of AI commercialization, beyond model capabilities, the importance of cost efficiency is rising rapidly.