蚂蚁国际开源AI预测大模型 超90%预测准确率+60%成本降幅

Core Insights - Ant International announced the open-source release of its AI forecasting model "Falcon TST" at the Singapore FinTech Festival 2025, marking it as the first large-scale time-series forecasting model based on a multi-segment pattern and a mixture of experts architecture, with over 2.5 billion parameters and optimal performance in various benchmark evaluations [1][3] Group 1 - The "Falcon TST" AI forecasting model was initially developed for internal use at Ant International for cash flow and foreign exchange risk prediction, achieving an accuracy rate exceeding 90% and potentially reducing foreign exchange costs by up to 60% [3] - The model can predict on an hourly, daily, or weekly basis and is applicable beyond finance, including weather changes, holiday consumption, financial market fluctuations, and cross-border human flow [3] - Ant International is collaborating with partners in industries such as aviation, banking, online travel, and e-commerce to explore specific applications of the model [3] Group 2 - In the aviation sector, the model can optimize foreign exchange hedging strategies, with pilot projects showing significant reductions in foreign exchange costs; it can also help reduce operational costs by 30% to 50% depending on the business model [3] - According to a report by the International Airports Council (ACI World), global air passenger volume is expected to reach 9.8 billion by September 2025, highlighting the importance of AI-driven precise forecasting for corporate profits and consumer benefits [3] - Ant International's Chief Innovation Officer stated that the decision to open-source the "Falcon TST" model aims to empower more industries and promote the iterative upgrade of AI technology in collaboration with academia and industry [3]