AT&T says open models now account for about 25% of its AI usage and could climb to 70–80%. In some applications switching from closed proprietary models to open models has cut costs by 80–90%. Goldman Sachs reports enterprises are adopting model tiering: lower-inference tasks shift to lower-cost open models while the most complex tasks remain on higher-capability proprietary models. As Chinese domestic open models increasingly substitute for U.S. models, Goldman-tracked AI-token costs have falle

AT&T says open models now account for about 25% of its AI usage and could climb to 70–80%. In some applications switching from closed proprietary models to open models has cut costs by 80–90%. Goldman Sachs reports enterprises are adopting model tiering: lower-inference tasks shift to lower-cost open models while the most complex tasks remain on higher-capability proprietary models. As Chinese domestic open models increasingly substitute for U.S. models, Goldman-tracked AI-token costs have fallen 49% since May, making model calls cheaper and supporting broader enterprise AI adoption. Goldman warns cheaper calls will pressure base-model vendors’ per-unit pricing and margins; cost efficiency is rising as a competitive axis alongside model capability in the next phase of AI commercialization.