Ant International's FalconTST 2.0 AI Sets New Forecasting Benchmark, Adopted by Major Banks

Ant International has unveiled a significant advancement in predictive artificial intelligence with the launch of its Falcon Time-Series Transformer (TST) Model 2.0. The upgraded model has secured a state-of-the-art (SOTA) position, achieving a leading score of 0.666 on the critical Mean Absolute Scaled Error (MASE) metric on a premier public benchmark for time-series foundational models. This technical milestone underscores the model's superior capability in analyzing sequential numerical data, a core requirement for dynamic financial environments where liquidity needs and currency positions fluctuate rapidly. The model's commercial validation is equally compelling, with adoption by several of the world's most prominent financial institutions. Barclays, Citi, Deutsche Bank, and Standard Chartered have integrated FalconTST 2.0 into their proprietary platforms to refine cash flow forecasting and foreign exchange management. This deployment has yielded a consistent forecast accuracy rate above 93%, a level of precision that directly enhances capital efficiency and risk mitigation for institutions processing vast volumes of international transactions. For instance, Barclays utilizes the model within its BARX NetFX hedging platform, while Citi has combined it with its Fixed FX Rates solution. Beyond banking, Ant International is positioning FalconTST as a reusable foundational capability for industries plagued by complex, multi-currency cash flows. The aviation sector, where revenues and costs span numerous currencies, is a primary use case. The company is actively expanding applications into e-commerce demand forecasting and logistics. Technically, the model's innovations include sophisticated handling of missing data, powerful generalization across domains like retail and energy, and native support for multiple time frequencies—from seconds to months—within a single architecture. "FalconTST is about understanding how the world changes over time and anticipating what comes next," said Jiang-Ming Yang, Chief Innovation Officer at Ant International. He emphasized that the value lies not merely in superior calculation but in transforming predictive intelligence into tangible decisions on liquidity and capital allocation. Kelvin Li, General Manager of Platform Tech, noted that the enhanced accuracy of version 2.0 allows the benefits of operational savings and confident cross-border transaction management to be extended to banking partners and a broader customer base. Ant International has made an API trial of FalconTST 2.0 available via GitHub to solicit developer feedback and accelerate innovation in time-series learning.