[Citigroup: AI Inference Efficiency Improves, Cutting-Edge Proprietary Model Task Costs Drop Significantly] According to a report by Golden Finance on September 28, Citigroup released a research report stating that the latest cutting-edge models have shown improvements in both performance and task costs, highlighting enhanced model efficiency and inference cost-effectiveness. This may help strengthen the long-term economic advantages of higher-efficiency models compared to lower-efficiency open-source weight alternatives. The report noted that the cost per task for open-source models dropped by 35% weekly to $0.8, while the discount relative to closed models widened from 40% to 60%. The proprietary model's intelligence lead increased from 9 points to 12 points.