weather
WeatherNext 2 (FGN)
Google weather ensemble model; beats WeatherNext Gen on 99.9% of variables and lead times (CRPS). Google docs give June 2025 as model-version date; WeatherNext 3 (Aug 2026) is now recommended.
Levelmodel
Classweather
Parametersnot in the notes
Contextnot in the notes
Released17 Nov 2025
Opennessproprietary
Runs onhyperscale cluster
Statusdeprecated
Benchmark Results
maintainer-run
WeatherBench 2 2domain
Z500 day 10, deterministic, FGN (oper.) (1st member) vs Analysismaintainer-run
811.7
0450900
Score in RMSE m2/s2. Lower is better.
year 2022, 240x121 grid
Z500 day 10, deterministic, FGN (oper.) (mean) vs Analysismaintainer-run
583.1
0450900
Score in RMSE m2/s2. Lower is better.
year 2022, 240x121 grid
Z500 day 5, deterministic, FGN (oper.) (1st member) vs Analysismaintainer-run
313.7
0450900
Score in RMSE m2/s2. Lower is better.
year 2022, 240x121 grid
Z500 day 5, deterministic, FGN (oper.) (mean) vs Analysismaintainer-run
233.1
0450900
Score in RMSE m2/s2. Lower is better.
year 2022, 240x121 grid
Z500 day 3, deterministic, FGN (oper.) (1st member) vs Analysismaintainer-run
139.4
0450900
Score in RMSE m2/s2. Lower is better.
year 2022, 240x121 grid
Z500 day 3, deterministic, FGN (oper.) (mean) vs Analysismaintainer-run
104.5
0450900
Score in RMSE m2/s2. Lower is better.
year 2022, 240x121 grid
Z500 day 1, deterministic, FGN (oper.) (1st member) vs Analysismaintainer-run
43.7
0450900
Score in RMSE m2/s2. Lower is better.
year 2022, 240x121 grid
Z500 day 1, deterministic, FGN (oper.) (mean) vs Analysismaintainer-run
32.3
0450900
Score in RMSE m2/s2. Lower is better.
year 2022, 240x121 grid
T850 day 10, deterministic, FGN (oper.) (1st member) vs Analysismaintainer-run
3.632
024
Score in RMSE K. Lower is better.
year 2022, 240x121 grid
T850 day 10, deterministic, FGN (oper.) (mean) vs Analysismaintainer-run
2.595
024
Score in RMSE K. Lower is better.
year 2022, 240x121 grid
T850 day 5, deterministic, FGN (oper.) (1st member) vs Analysismaintainer-run
1.81
024
Score in RMSE K. Lower is better.
year 2022, 240x121 grid
T850 day 5, deterministic, FGN (oper.) (mean) vs Analysismaintainer-run
1.348
024
Score in RMSE K. Lower is better.
year 2022, 240x121 grid
Where this comes from
Every score names who ran it: vendor-reported, maintainer-run or independent. A rank is shown with the population it was ranked in (gap rule 3).
Lineage
successor
GenCast (WeatherNext 1 Gen) is the parent (successor)high confidencenotesGraphCast (WeatherNext 1 Graph) is the parent (successor)high confidencenotesWeatherNext 3 is a child (successor)high confidencenotesWhere this comes from
Parent and child relations from model cards, Hugging Face base_model fields and the genealogy tables of the research notes. The tree shows two generations each way, 6 models. Confidence and evidence are the notes' own.
Industry Use Cases
Agriculture
Weather forecastingR&D Topics
Identity
- Developer
- Google DeepMind, Google
- Family
- not in the notes
- Released
- 17 Nov 2025 sourcesannouncement page
- Weights published
- –
- Record checked
- 10 Oct 2026
Where this comes from
Developer and dates from the research notes. Release date is the vendor's public announcement; weights date is when open weights were published.
Architecture and Size
- Architecture
- other Functional Generative Network (noise injected into architecture), ensemble; hourly output
- Total parameters
- – sources
- Active parameters
- –
- Checkpoint parameters
- –
- Note
- Google docs give June 2025 as the model-version date
- Context length
- not in the notes
- Input
- sensor
- Output
- sensor
Where this comes from
Parameter counts as the developer states them. Where the checkpoint holds a different total, both are shown (register A16).
Openness and Access
- Openness
- proprietary
- Licence
- Proprietary sources
- Access routes
- cloud marketplace, gated preview
Where this comes from
The licence file governs over registry tags and aggregator labels.
Deployment
- Runs on
- hyperscale cluster
- Footprint
- Hundreds of members in under 1 min on one TPU
- Runtimes
- not in the notes
Where this comes from
Deployment tiers and runtimes as named in the notes. Footprint is the notes' figure, not a measurement by AURION.
Indonesia
- Indonesian support
- not stated
- Regional languages
- not in the notes
- Jakarta hosting
- unknown
- Languages
- not in the notes
Where this comes from
Indonesia fields are carried on every model; unknown is an allowed value.
Links
ModelWeatherNext 2 (FGN)Google DeepMind, weather
WeatherNext 2 (FGN)Google DeepMind, weatherUse CasesWeather ForecastingAgriculture
Who Runs It
Badan Meteorologi, Klimatologi, dan Geofisika (BMKG)Contract signed, 5 Feb 2026Badan Meteorologi, Klimatologi, dan Geofisika (BMKG)In production, 23 Sep 2025Deployments of these use cases; none names this model.
Badan Meteorologi, Klimatologi, dan Geofisika (BMKG)Contract signed, 5 Feb 2026Badan Meteorologi, Klimatologi, dan Geofisika (BMKG)In production, 23 Sep 2025Deployments of these use cases; none names this model.Use cases 1
Models 3
Sources
- WeatherNext 2: Our most advanced weather forecasting model blog.google/technology/google-deepmind/weathernext-2 https://blog.google/technology/google-deepmind/weathernext-2/
- developers.google.com/weathernext/guides/evals https://developers.google.com/weathernext/guides/evals
- WeatherNext 3 | Google for Developers developers.google.com https://developers.google.com/weathernext/guides/models
- WeatherNext 2 | Google for Developers developers.google.com https://developers.google.com/weathernext/guides/models-wn2
- Model deprecation | WeatherNext | Google for Developers developers.google.com https://developers.google.com/weathernext/guides/deprecation
- WeatherNext 2: Our most advanced weather forecasting model blog.google/technology/google-deepmind/weathernext-2 https://blog.google/technology/google-deepmind/weathernext-2/
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "WeatherNext 2 (FGN)", https://aurion.semai.id/models/google-deepmind-weathernext-2/, records of October 2026.