Google Research releases TimesFM-3, a multivariate time-series forecasting model

Google Research released TimesFM-3, a 330-million-parameter open-weights forecasting model available on GitHub and Hugging Face. Unlike the prior TimesFM-2.5, it handles multiple related time series at once, can factor in known future events like planned discounts or weather, and outputs a full forecast in one pass rather than step by step. Google reports it topping three benchmarks (Gift-Eval, FEV-Bench, Time) against Amazon's Chronos-2 and the Toto-2.0 family. A BigQuery integration is promised "in the coming weeks" but not yet live.

This is a real, usable artefact, not a paper or a demo: open weights mean developers and enterprises can test it today on retail, energy, or operations forecasting problems where multiple correlated signals matter. For teams already using TimesFM or Chronos, multivariate support and single-pass generation are a genuine engineering improvement, likely cheaper and faster inference with fewer compounding errors. The impact is bounded, though: this is a niche technical upgrade for forecasting teams, not a shift that changes who holds power in AI, and enterprise-grade adoption depends on the still-unshipped BigQuery integration.

Every benchmark comparison comes from Google, run on benchmarks and against competitors Google chose, so "ranks first" should be read as a vendor claim until someone reproduces it independently. The coverage frames this as predicting "the future from sales data, weather, and discount schedules," which oversells a fairly standard time-series forecasting upgrade as something more dramatic. Calling this a step change is misleading. It is an incremental architecture improvement over TimesFM-2.5 and a competitive response to Chronos-2, not a new category of capability.

1. 01Independent third-party benchmark results on Gift-Eval or FEV-Bench comparing TimesFM-3 against Chronos-2 and Toto-2.0, outside Google's own write-up
2. 02Whether the promised BigQuery integration actually ships in the coming weeks, and on what terms (free tier, pricing, GA vs preview)
3. 03Developer adoption signals: GitHub stars/forks trajectory, Hugging Face download counts, and whether enterprises publicly report switching from Chronos-2 or TimesFM-2.5