Ant International Open-Sources Falcon TST AI Model for Time-Series Forecasting

Ant International has open-sourced its proprietary Falcon TST (Time-Series Transformer) AI model. The model is based on a Mixture of Experts architecture and includes multiple patch tokenizers, supporting up to 2.5 billion parameters.

Falcon TST has achieved state-of-the-art zero-shot performance on long-term forecasting benchmarks, with internal deployment at Ant International showing over 90% accuracy in managing cashflow and foreign exchange exposure. The company reported that the model reduced FX costs by as much as 60%.

The AI system has been used with industry partners for applications such as mitigating FX volatility and helping airlines stabilize pricing. It can also be applied to other time-series forecasting tasks, including weather prediction, financial market analysis, and cross-border traffic monitoring.

The Falcon TST model is now publicly available on GitHub and Hugging Face, with Ant International inviting global collaboration to advance research in time-series learning.