Google released TimesFM-3, a zero-shot foundation model designed for multivariate time-series forecasting. It predicts multiple interdependent variables simultaneously without requiring task-specific training. This architecture reduces the need for custom model tuning in complex data environments.