Ant International released version 2.0 of FalconTST, its model for forecasting cash flow and foreign exchange exposure, on August 20, 2026, and named the banks running it in production: Barclays, Citi, Deutsche Bank, and Standard Chartered. The company says the new version delivers forecast accuracy above 93% at its partner banks and ranks first on a public benchmark leaderboard for the error metric it reports.
A transformer trained on numbers, not words
TST stands for Time-Series Transformer, an architecture trained on sequences of time-stamped numerical values. FalconTST learns patterns from series drawn from finance, retail, energy, travel, and economic statistics. A language model learns relationships between words; a time-series model learns relationships between successive quantities such as liquidity needs, rate movements, and transaction volumes. “While large language models excel at learning relationships in text, TST models are especially critical in finance and payments, where liquidity needs, foreign-exchange movements, and transaction flows can shift rapidly,” Ant International said, as quoted by PYMNTS. In payments, that makes the forecasting window short and expensive to miss.
Ant says the model helps businesses know “precisely when they need funds, how much they need, and in which currencies.” Each of those questions has a direct treasury consequence:
- When funds will be needed, down to the hour and the value date.
- How much capital has to be set aside to meet expected settlements.
- Which currencies that capital must be held in, which determines the hedging required.
Forecasting errors cost providers idle cash or rushed hedges
A payment that crosses a currency border needs funds already sitting in the destination currency. The provider therefore prefunds local accounts, ties up cash, and carries the risk that the rate moves between the moment it quotes and the moment it settles. Every hour of advance notice on that need shrinks the buffer it has to hold, and every forecasting error is paid for either in idle liquidity or in hedging bought at the last minute.
Currency conversion has long been a profit line in payments. Card networks Visa and Mastercard publish their own conversion rates, and issuers add a fee on top. A forecasting model works further upstream, on the volume of hedging needed to honor a quoted rate.
Each bank plugs the model into its own FX stack
| Bank | Integration point |
|---|---|
| Barclays | BARX NetFX FX platform |
| Citi | Fixed FX Rates product |
| Deutsche Bank | FX operations |
| Standard Chartered | SCALE FX system, under the Monetary Authority of Singapore’s PathFin.ai program |
Kelvin Li, general manager of platform tech and senior vice president at Ant International, pitched version 2.0 as a way to reach new users: “With FalconTST 2.0, enhanced accuracy and precision let us extend those benefits to our banking partners as well as a broader range of customers across fast-moving sectors like e-commerce, travel and fintech.” Chief Innovation Officer Jiang-Ming Yang focused on how the forecasts are used: “the value of AI is not simply achieving a better forecasting score, but turning that predictive intelligence into real decisions—how much liquidity to prepare, how to manage FX exposure, and how to allocate capital more efficiently.”
The 93% accuracy claim comes without its terms
Adoption by four large international banks is a different kind of validation from a leaderboard ranking. A bank that plugs an outside model into a live FX platform agrees to let hedging decisions depend on it, with the internal controls and model governance obligations that come with that.
A payments provider sells its forecasting tool to banks
Ant International also runs its own cross-border payment rails and serves merchants in the same corridors as its bank customers. Selling the forecasting tool to four banks means monetizing an internal component with firms that remain competitors in merchant acceptance. Ant says it plans to extend the model to supply chain demand forecasting and to predictive operations in aviation.
For European payment providers, the deal turns a cost question into an ownership question. FX margin depends on forecast quality as much as on the quoted rate, and forecasting is becoming a component that firms either buy or build. Buying means depending on a vendor that processes payments itself.