Google releases TimesFM-3, a zero-shot multivariate time series foundation model
rseroter · x · 2026-09-02
Google Research introduced TimesFM-3, a next-gen time series foundation model with 330M parameters pre-trained on over 1 trillion time points. Its key breakthrough is native support for multivariate forecasting. It predicts multiple co-evolving series using historical data and auxiliary features (e.g., weather) in a zero-shot manner, outperforming baselines on major benchmarks. It is available on GitHub and Hugging Face.
More from Models
- Anthropic launches Claude Fable 5.1: 2x jump on Terminal-Bench-Science, 25% cheaper — heathercmiller · 2026-09-02
- User finds regular Claude web sessions appear to run in a Linux VM — majidmanzarpour · 2026-09-02
- Perplexity uses Fable 5.1 as orchestrator with GPT 5.6 as cost-efficient subagents — AravSrinivas · 2026-09-02
- Fable 5.1 leaps ahead on price-performance, beating Sol 5.6 Max on Cursor Bench for less — kimmonismus · 2026-09-02
- Fable 5.1 Hands-On: Stronger Than Fable 5, Half the Tokens of Opus 5 — every · 2026-09-02
- Fable 5.1 takes 1st on Artificial Analysis with a score of 66 — Anxious-Yoghurt-9207 · 2026-09-02