Ant International open-sources Falcon AI model for forecasting

Ant International has released its Falcon TST AI model, which it uses internally to forecast cashflow and foreign exchange (FX) exposure.

The Singapore-headquartered fintech firm said the model is built with a mixture of experts architecture, incorporates multiple patch tokenizers, and uses up to 2.5 billion parameters.

According to Ant International, the Falcon TST model has reached over 90% accuracy for cashflow and FX exposure forecasting, and helped reduce its FX costs by up to 60%.

The company said it has collaborated with industry partners to apply the model in areas such as helping businesses manage FX volatility, and supporting more stable, competitive airline pricing for customers.

Ant International has made the Falcon TST model available as open source on GitHub and Hugging Face.

The company said the model can be used for various time-series forecasting tasks, including weather prediction, financial market trends, and cross-border traffic data.

🔗 Source: Ant International

🧠 Food for thought

Implications, context, and why it matters.

Claims of 90%+ accuracy lack disclosed evaluation details to assess enterprise readiness

Ant International says its forecasts hit over 90% accuracy and cut FX costs by up to 60%. It gives no details on datasets or benchmarks. It also omits metrics such as mean absolute error [MAE], root mean square error [RMSE] and mean absolute percentage error [MAPE]. There are no baseline comparisons for independent checks.
The model is listed on GitHub and Hugging Face. The post does not say if the model weights (the learned parameters) are available or which open-source license applies. It also does not say whether training code or datasets are included, which blocks reproducibility and buyer checks.
A mixture of experts architecture (a design that routes inputs to specialized sub-models) with 2.5 billion parameters implies heavy compute needs. There is no data on inference costs, latency, or hardware. Small and medium-sized businesses (SMBs) cannot judge if real-time FX risk management is viable.

Third-party developers can build managed services around Falcon TST for the underserved SMB FX risk market

Falcon TST is open-sourced for time-series forecasting (predicting future values from historical time-stamped data). Independent software vendors can offer managed hosting, fine-tuning, and integrations with enterprise resource planning (ERP) systems and accounting platforms. Small and medium-sized businesses (SMBs) could forecast cashflow and manage FX exposure without building in-house AI.
SMB software market reached USD 72.35 billion in 2025 and is growing at 6.98% compound annual growth rate (CAGR). Cloud deployment leads the segment, and about 63% of SMB workloads run in the cloud 1. That base supports adoption of new software-as-a-service (SaaS) financial tools.
Financial risk management software is set to grow from USD 4.28 billion in 2025 to USD 14.39 billion by 2034 at 14.42% CAGR 2. Small and medium-sized enterprises (SMEs) are expected to grow fastest, and high solution costs remain a restraint. Lower priced offerings built on Falcon TST could fit this need.