Google Open-Sources WeatherNext 2: Generates 15-Day Forecasts on a Single TPU in Under a Minute
aigclink · x · 2026-08-07
Google has open-sourced its most advanced AI weather forecasting model, WeatherNext 2, along with a dedicated cyclone version and a mini version capable of running on a single TPU.
Surpassing its predecessor in both forecast lead time and meteorological variables, the model features:
- Ultra-fast inference: Completes a global 0–15 day forecast on a single TPU in under 1 minute, an 8x speedup.
- High temporal resolution: Achieves 1-hour granularity and outputs hundreds of probabilistic scenarios per run, enabling near real-time dynamic forecasting.
- Complex correlation prediction: Trained on single-point weather elements, it can predict large-scale joint systems, solving complex correlation issues that traditional models struggle with.
The model's data is integrated into Earth Engine and BigQuery for developer queries, and Google Cloud Vertex AI has opened early access supporting custom inference.
Related event: Google Open-Sources WeatherNext AI for Cyclone Forecasting(9 posts)→
More from Models
- ByteDance Training 10-Trillion Parameter Model to Rival Anthropic — Ars Technica AI · 2026-08-07
- Users Praise DeepSeek's CoT: A 'Whale Thought' Experience Battling Its Own Slop — teortaxesTex · 2026-08-07
- Testing NVIDIA Nemotron 3.5 ASR: Blazing Fast but Misses Crucial Drug Names — MaziyarPanahi · 2026-08-07
- LiquidAI LFM2.5-2.6B Quantization Report: Runs on Raspberry Pi — crusaderky · 2026-08-07
- Polymarket Predicts Alibaba as Top Chinese AI Firm with 89% Odds — Polymarket · 2026-08-07
- Professor's Test: Gemini Shows More 'Insight' Than ChatGPT — ubizmo · 2026-08-07