---
format: "aidr-story-markdown/v1"
id: "565cd36c470667a3cd043984e76e75876011fd41f923f4e2b516a34646716866"
canonical_url: "https://aidr.today/565cd36c?lang=en"
title: "Google DeepMind WeatherNext 3 Hits 5 km Resolution For First Hourly Global AI Forecasts"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-09-06T18:47:01.000Z"
category: "Research"
topics: ["google","deepmind","multimodal","vision-language","benchmark","inference"]
source_urls: ["https://huggingnews.com/ai/google-deepmind-weathernext-3-hits-5-km-resolution-for-first-hourly-glob-9dfbf3e1","https://x.com/dair_ai/status/2096644646780973297","https://x.com/JaminurAI/status/2096485028524097591","https://x.com/TwelveCacti/status/2096359973630419054"]
summary: "Google DeepMind integrated its WeatherNext 3 model into Search, Maps and Gemini to provide global weather updates every hour. The model operates at a 5 km resolution, or 0.1 degree, whereas previous AI models typically worked over patches of 15 to 25 square kilometers. On benchmarks cited by Google, the system outperformed AI models from Microsoft and Nvidia as well as traditional forecasts from the National Weather Service. The model shifts training from traditional analysis data—which are outputs of other models—to raw low-latency geostationary satellite data. By learning targets in observation space, WeatherNext 3 predicts satellite-derived precipitation and tropical cyclone tracks directly. This approach reduces inherited biases from previous models and produces lower errors for 2m temperature and dewpoint at specific locations."
---

# Google DeepMind WeatherNext 3 Hits 5 km Resolution For First Hourly Global AI Forecasts

> [Open the canonical story](<https://aidr.today/565cd36c?lang=en>)

**Published:** 2026-09-06T18:47:01.000Z
**Category:** Research
**Topics:** google, deepmind, multimodal, vision\-language, benchmark, inference

## Summary

Google DeepMind integrated its WeatherNext 3 model into Search, Maps and Gemini to provide global weather updates every hour\. The model operates at a 5 km resolution, or 0\.1 degree, whereas previous AI models typically worked over patches of 15 to 25 square kilometers\. On benchmarks cited by Google, the system outperformed AI models from Microsoft and Nvidia as well as traditional forecasts from the National Weather Service\. The model shifts training from traditional analysis data—which are outputs of other models—to raw low\-latency geostationary satellite data\. By learning targets in observation space, WeatherNext 3 predicts satellite\-derived precipitation and tropical cyclone tracks directly\. This approach reduces inherited biases from previous models and produces lower errors for 2m temperature and dewpoint at specific locations\.

## Sources

- [Story source](<https://huggingnews.com/ai/google-deepmind-weathernext-3-hits-5-km-resolution-for-first-hourly-glob-9dfbf3e1>)
- [Supporting source](<https://x.com/dair_ai/status/2096644646780973297>)
- [Supporting source](<https://x.com/JaminurAI/status/2096485028524097591>)
- [Supporting source](<https://x.com/TwelveCacti/status/2096359973630419054>)