HONG KONG, August 20, 2026, 14:14 HKT
- Ant International rolled out its Falcon Time-Series Transformer Model 2.0. Six banks partnered on the launch.
- The company says using precise forecasts can lower FX hedging and allocation costs by over 60%.
- Performance results are reported by Ant and its partners, without an independent audit.
Ant International rolled out its Falcon Time-Series Transformer Model 2.0 on Thursday. Six major banks are working with Ant on the new forecasting tool. The company said the forecasts can cut foreign-exchange hedging and allocation costs by over 60%.
The model is built for a specific costly problem. Treasurers at companies have to estimate cash requirements and FX exposure ahead of hedging. If they predict wrong, they could hedge too much or miss some risks.
Citigroup (NYSE:C), HSBC Holdings (LON:HSBA), Deutsche Bank (ETR:DBK), Standard Chartered (LON:STAN), and Barclays (LON:BARC) are named as partners. Ant left the sixth bank unnamed in the Reuters announcement.
| Model measure | Current public Falcon profile | Falcon TST 2.0 launch disclosure |
|---|---|---|
| Primary task | Time-series forecasting | Financial forecasting and liquidity risk |
| Architecture | Mixture of Experts and several patch tokenisers | Not given separately |
| Parameters | 2.5 billion | Not given separately |
| Training time points | 300 billion | Not given separately |
| Hourly FX-demand accuracy | Above 90%, company’s number | Updated outside result not given |
Falcon’s product site shows 2.5 billion parameters and 300 billion time points, using a Mixture of Experts setup. Ant says it gets over 90% accuracy for hourly FX-demand forecasts.
| Bank partner | Verified earlier Falcon use | 2.0 status |
|---|---|---|
| Citigroup (NYSE:C) | Pilot for Fixed FX Rates serving airline accounts | Named partner |
| Barclays (LON:BARC) | BARX NetFX integration in place | Named partner |
| Standard Chartered (LON:STAN) | Integrated with SCALE for round-the-clock treasury tools | Named partner |
| Deutsche Bank (ETR:DBK) | Falcon looked at for cross-border treasury work | Named partner |
| HSBC Holdings (LON:HSBA) | No details on earlier Falcon rollouts | Named partner |
| Sixth bank | No disclosure | Name not disclosed |
Instead of a lab project, the rollout builds off live banking work. Citi combined Falcon with a fixed-rate product that handles 70-plus currencies. Barclays and Standard Chartered plugged the model into their FX systems.
“Accurate forecasting can cut FX hedging and allocation costs by more than 60%,” Kelvin Li, general manager of platform tech at Ant International, told Reuters.
| Reported deployment | Performance or saving | Scope and source |
|---|---|---|
| Ant International FX forecasting | Accuracy above 90% | Forecasts by hour, day and week; company says |
| Ant FX-conversion flow | Over 60% of the matching transactions | Listed in 2025 partner release |
| Ant and customer FX costs | Reduction as much as 60% | Listed in 2025 partner release |
| Liquidity-management costs | Reduction up to 50% | Listed in 2025 partner release |
| Citi airline pilot | Hedging costs cut by 30% | First live transactions |
| Capital A (KLSE:5099) | Hedging costs down 40% | Live use |
The savings can’t be lined up side by side. They include different clients, time periods and types of treasury products. Standard Chartered saw FX costs fall by as much as 60% and liquidity-management costs by up to 50% with earlier Ant use.
Citi’s airline pilot delivered a 30% cut in hedging costs in first live trades. Capital A logged a 40% drop. Both results set a more modest range than Thursday’s wider claim.
Falcon puts out exposure forecasts, but doesn’t replace the bank’s execution tools. Banks keep providing liquidity, handle rate locks, and settle deals. The model shifts how much protection is asked for, and when.
Ant International pulled in $1.2 billion last month as it looks to grow beyond China. The Singapore-based company is Ant Group’s international arm.
Risks: Ant and partners provided the reported performance numbers. Results could shift with FX moves, market swings and client information. For regulated treasury, issues like model checks, privacy, and human review are still key.
Specialist AI is expanding into mainstream finance with a six-bank rollout. The commercial test is clear—now it’s about seeing if the wider rollout can match the cost savings shown in smaller pilots.
