Ant International said it has open-sourced its Falcon Time-Series Transformer (TST), an AI model it uses to forecast cash flow and foreign-exchange (FX) exposure, as the payments and fintech firm seeks broader adoption of AI-driven forecasting across industries.
The company said Falcon TST uses a Mixture-of-Experts architecture with up to 2.5 billion parameters and has delivered accuracy rates above 90% in internal deployment, helping cut FX costs, though it did not disclose training data, baseline comparisons or full evaluation methodology.
Ant has been pitching the technology to corporates over the past year. Malaysia’s Capital A, parent of AirAsia, said in October that Ant’s AI forecasting helped reduce its FX hedging costs by as much as 40% and enabled more stable ticket pricing, an example of how airlines could use predictive models to smooth revenue and currency risk.
The model’s release comes as banks and treasury teams test AI for risk management and liquidity planning.
Citigroup and Ant announced a pilot in July to use Falcon-based tools to help clients manage FX risk, reflecting rising interest among financial institutions in machine-learning approaches to time-series forecasting.
Ant said Falcon TST has shown state-of-the-art “zero-shot” results on long-horizon benchmarks and can be adapted for other time-series tasks, from traffic and calendar effects to commodity and weather-linked forecasting.
The company added that code and model artifacts are available on GitHub and Hugging Face; Reuters could not immediately verify the repositories’ contents. (Ant Group has previously maintained open-source projects.)
Airlines are a prime target market: Airports Council International World projects 2025 global passenger traffic at about 9.8 billion, near the pre-pandemic peak, intensifying the need for sharper demand and pricing forecasts.
Stable fare strategies linked to better FX and sales projections could aid carriers as cross-border traffic continues to recover unevenly.
Ant did not disclose licensing terms beyond saying the model is “open-sourced.” The firm invited external developers to test and provide feedback on real-world use cases spanning finance, logistics and travel.
Business News Asia
