Google releases TimesFM-3: zero-shot multivariate time-series forecasting in a single forward pass
ymatias · x · 2026-09-01
Google Research has released TimesFM-3, the third generation of its time-series foundation model, now open on GitHub and Hugging Face, with BigQuery integration coming soon.
- Native multivariate support: unlike earlier versions (through TimesFM-2.5) limited to univariate forecasting, TimesFM-3 natively handles multivariate series and auxiliary external features, jointly predicting multiple coevolving series zero-shot in one forward pass.
- Scale: 330M parameters, pre-trained on more than 1 trillion time points from real-world and synthetic corpora.
- Results: ranks first among all pretrained foundation models on major public benchmarks including Gift-Eval, FEV-Bench and Time.
The team notes real-world forecasting (e.g., retail ice cream sales) is inherently multivariate, requiring related product sales, foot traffic, weather and promotions as signals.
More from Models
- OpenAI stealth-testing mysterious internal model 'Aeon' in Codex desktop app and Dots — cedric_chee · 2026-10-16
- Argon edges out astra as daily driver, but Claude still wins heavy coding work — iruletheworldmo · 2026-10-16
- NVIDIA ships FP4-quantized Qwen3.8-Flash for local use, trending on Hugging Face — nvidia · 2026-10-16
- RoboVista benchmark: Gemini-4 Argon hits 81.9% on robotics understanding — berkeley_ai · 2026-10-16
- Dev reports decision model jev beats gpt-6-luna on quality in real training pipeline — genmon · 2026-10-16
- Google won't publish Argon 4 limits in AI Pro—users compare experiences — Consistent-Text-2662 · 2026-10-16