FalconTST 2.0 moves predictive AI into bank foreign-exchange workflows
Ant International released Falcon Time-Series Transformer 2.0 and said six global banks, including Citi, HSBC, Deutsche Bank, Standard Chartered and Barclays, are using FalconTST technology in foreign-exchange or treasury workflows. The open implementation is available on GitHub. Ant reports a 0.666 MASE result on GIFT-Eval and more than 93% forecast accuracy; these metrics are vendor-reported.
Why it matters: Specialised time-series foundation models are moving from internal forecasting into regulated financial products and treasury operations.
Business/technical/strategic impact: Banks should validate against their currencies, horizons and regime shifts, preserve deterministic risk limits and distinguish forecast quality from executable trading performance. Pricing for production use was not disclosed.
