• •Google Research released TimesFM-3, featuring 330 million parameters and support for multivariate forecasting.
• •MarkTechPost confirmed the model handles multiple related data series in a single forward pass, a departure from prior versions.
• •Documentation from Google Research highlights the model’s 'zero-shot' foundation capabilities, though benchmarks against industry-standard specialized models remain sparse.

Google Research has launched TimesFM-3, a 330-million parameter model designed to handle multivariate time series forecasting in a single forward pass. While MarkTechPost emphasizes the technical shift from the previous 2.5 iteration, official Google documentation focuses primarily on the model's zero-shot foundation architecture. Neither source provides independent comparative benchmarking against established statistical methods, leaving the model's practical efficacy in production environments uncertain. Determining whether this foundation approach outperforms legacy algorithms will likely require broader community testing.