Ant International’s FalconTST 2.0 Now Powers FX at Four Global Banks

Time-series forecasting is the quiet workhorse of cross-border finance, and Ant International — the Singapore-headquartered arm of Ant Group — is betting the transformer architecture that reshaped language AI can do the same for predicting cash flow and currency swings. Its newest model, FalconTST 2.0, is now embedded in the FX hedging operations of four of the world's largest banks, a practical signal that applied AI is moving from generating text to anticipating what happens next in global markets.

FalconTST 2.0 is the updated version of Ant's Falcon Time-Series Transformer, an AI model built around the finding that financial, energy and retail data share common temporal structures — cycles, trends, sudden shifts and seasonality. The model scored 0.666 on Mean Absolute Scaled Error (MASE), a benchmark metric for time-series accuracy that Ant says outperforms other transformer foundation models from leading global tech firms. First deployed internally to manage cash flow and FX exposure on hourly, daily and weekly cycles, it has since been taken up by Barclays, Citi, Deutsche Bank and Standard Chartered, which have integrated it into their hedging models to improve cash-flow forecasting and FX liquidity management.

FalconTST learns one set of patterns across diverse industries rather than building a separate model per task, which is why Ant sees a wider runway ahead: demand forecasting for e-commerce supply chains and predictive operations management for aviation are flagged as the next applications. It fits a broader trend in which companies are discovering that the forecasting layer of business — treasury, inventory, route planning — has been largely untouched by the recent wave of generative AI, and is now where the returns are showing up first.

For bankers and finance teams across Singapore and Asia, the significance is that four global institutions are running mission-critical FX hedging on a model developed by an Ant Group unit headquartered here. That is a real-world vote of confidence in Singapore as a place where frontier financial-technology models are built and shipped globally, not merely hosted.

Why it matters for Singapore: Ant International's Singapore base means the city-state is not just a regional headquarters for global finance but increasingly a source of the AI models those institutions run on. As time-series AI moves into treasury management, supply-chain planning and aviation, FalconTST 2.0 offers a concrete reminder that the most valuable AI work coming out of Singapore is often the invisible kind — powering the forecasts and hedging decisions that keep cross-border money moving.