At present, enterprise AI applications have stepped into a novel phase characterized by model differentiation. Owing to the steep pricing of closed-source, cutting-edge models from companies such as OpenAI and Anthropic, a significant number of US enterprises are now redirecting non-core AI tasks—those with well-defined fault tolerance limits—towards open-weight models, primarily as a cost-saving measure. Many firms have witnessed their annual AI expenses skyrocket tenfold within a mere six-month span. Industry behemoths like AT&T, which process hundreds of billions of Token requests daily, have already transitioned 40% of their AI workloads to open models and aim to elevate this figure to 70% within the coming year. Following fine-tuning with proprietary datasets, open models can deliver performance on par with, or even outperform, their closed-source counterparts in specific tasks. Over the past year, both the frequency of open model mentions among enterprises and the actual volume of Token requests have surged notably. Chinese companies, including DeepSeek and Zhipu, have emerged as pivotal forces in the global open model landscape, capitalizing on the US enterprises' drive to slash AI costs. Beyond cost benefits, open models also cater to enterprise requirements concerning data sovereignty, security, and bespoke customization. Presently, companies have not entirely forsaken closed-source, cutting-edge models, as intricate and demanding tasks still necessitate their use. Nevertheless, enterprises are generally starting to bifurcate AI tasks and implement differentiated model routing strategies, with cost considerations becoming a pivotal factor shaping the market share of leading closed-source vendors.