Toto AI Model Sets New Benchmark for Time Series Forecasting

Toto, a pre-trained time series foundation model, delivers state-of-the-art performance across Long Sequence Forecasting (LSF) and Datadog benchmarks. Leveraging its Proportional Factorized Space-Time Attention mechanism and massive multi-domain training set of one trillion time points, Toto outperforms leading zero-shot models like Moirai, TimesFM, and Chronos on key metrics such as MAE, MSE, and sMAPE. Its strength lies in capturing complex temporal-spatial dependencies and adapting to diverse data conditions, making it a standout performer for both general-purpose forecasting and observability time series prediction.