Google released TimesFM-3, a 330M-parameter time series foundation model with native multivariate forecasting (non-commercial license)

Explain Like I'm Normal

Google Research dropped TimesFM-3, a 330M-parameter model designed to predict future data trends without needing custom fine-tuning. Unlike previous versions, it handles 'multivariate' data, meaning it can factor in external variables like holidays or pricing changes simultaneously to predict outcomes. It uses a clever transformer architecture that looks both across time and across different data categories in a single pass.