Google releases TimesFM-3, a 330M-parameter model that forecasts time series in a single pass
The Decoder · rss · 2026-09-12
Google Research released TimesFM-3, a 330-million-parameter forecasting model that analyzes time series alongside related data and known future events such as sales promotions or weather forecasts.
Instead of predicting the future step by step, it fills in all future time points in a single pass, cutting compute time and reducing compounding errors.
More from Research
- 24 Fields Medalists Sign Protest Letter to OpenAI as Mathematicians Fear Talent Pipeline Collapse — ducha_aiki · 2026-09-12
- sqlite-sparse runs learned sparse retrieval inside SQLite with no model at query time — Much-Ingenuity5415 · 2026-09-12
- Lost in the Middle: LLMs Miss Facts Buried in Long Contexts, Study Shows — thisdudelikesAI · 2026-09-12
- GPT-3 to Chinchilla: A Thread Walking Through the Papers That Reshaped AI — thisdudelikesAI · 2026-09-12
- Kaplan's 2020 Scaling Laws: The Paper Behind Every 'Just Make It Bigger' Decision — thisdudelikesAI · 2026-09-12
- LiFlow: Lidar Point Cloud Completion Method Accepted to ECCV 2026 — ducha_aiki · 2026-09-12