Skip to content
ContentLora

    Tip: press / anywhere to search.

    product

    WeatherNext (GraphCast, GenCast)

    Also known as GraphCast, GenCast, WeatherNext 2, Functional Generative Network

    Google DeepMind's machine-learning weather models are now branded WeatherNext. GraphCast (2023) beat ECMWF's flagship deterministic forecast on most measures, GenCast (2024) beat its ensemble, and WeatherNext 2 (2025) improved on both in speed and skill.[1][2][3]

    Editor reviewedUpdated AI for scienceArtificial intelligenceScience
    Key facts

    WeatherNext is the name of Google DeepMind‘s current AI weather models; this page also covers the research models that preceded WeatherNext 2. Two of them, GraphCast and GenCast, beat the European Centre for Medium-Range Weather Forecasts (ECMWF) on its own deterministic and ensemble benchmarks.[1][2]

    GraphCast (2023)

    GraphCast is a graph neural network trained on four decades of ECMWF’s ERA5 reanalysis.[4] It takes the current weather and the weather six hours earlier, predicts six hours ahead, and repeats to reach 10 days.[5] In the Science paper, it beat ECMWF’s HRES forecast on more than 90% of 1,380 variables and lead times, and a 10-day forecast took under a minute on one TPU v4 machine.[1][6]

    GenCast (2024)

    Single forecasts cannot show uncertainty, and earlier ML models lagged physics-based ensembles on reliability.[7] GenCast, announced by Google DeepMind in December 2024, generates ensembles of 15-day global forecasts at 0.25° resolution for more than 80 variables in 8 minutes.[8][9] It beat ECMWF’s ENS ensemble on 97.2% of 1,320 targets and better predicted extreme weather, cyclone tracks and wind power.[2]

    WeatherNext 2 (2025)

    WeatherNext 2, announced on 17 November 2025, uses a Functional Generative Network that injects noise directly into the model so its scenarios stay physically consistent.[3] It produces hundreds of scenarios from one starting point, each in under a minute on a TPU, and beats the previous WeatherNext model on 99.9% of variables and lead times out to 15 days.[10][3]

    Cyclones and Weather Lab

    In June 2025 Google DeepMind and Google Research launched Weather Lab, a website for their AI weather models. Its experimental cyclone model predicts a storm’s formation, track, intensity, size and shape, generating 50 possible scenarios up to 15 days ahead.[11] Google said internal testing found its track and intensity predictions as accurate as, and often more accurate than, physics-based methods.[12] Forecasters at the US National Hurricane Center began seeing its live predictions alongside physics-based models and observations.[11] Weather Lab shows the AI forecasts next to ECMWF’s physics-based models, and Google stresses that it is a research tool whose predictions are not official warnings.[13][14]

    Access

    Google DeepMind’s public WeatherNext repository holds code for WeatherNext 2 and for GraphCast and GenCast. Daily WeatherNext 2 forecasts are offered through Google Cloud, Weather Lab and Open-Meteo.[15]

    Caveats

    The headline comparisons come from the developer’s own papers and blog posts.[1][3] The models also learn from ERA5, which is produced with a physics-based forecasting system, so they build on traditional meteorology rather than replace it.[4]

    Wider adoption

    Weather agencies have since built their own operational AI systems. ECMWF’s ecmwf-aifs went operational in February 2025, and NOAA deployed AI-driven global models in December 2025.[16][17] ECMWF publishes the AIFS weights under an open CC BY 4.0 licence.[18] Storm intensity is a known weak spot for the AIFS, whose model card lists reduced intensity of some tropical cyclones, and it is one of the measures on which Google evaluates its own cyclone model against physics-based methods.[19][12]

    Questions readers ask

    What data are WeatherNext models trained on?

    GraphCast was trained on four decades of ECMWF's ERA5 reanalysis, a reconstruction of past weather built from observations and a physics model.[4]

    What is the difference between GraphCast and GenCast?

    GraphCast produces a single 10-day forecast; GenCast generates an ensemble of 15-day forecasts to represent uncertainty.[5][9]

    How fast is WeatherNext 2?

    It produces each of hundreds of possible scenarios in under a minute on a single TPU.[10]

    Do hurricane forecasters use WeatherNext?

    Since June 2025 forecasters at the US National Hurricane Center have seen live predictions from Google DeepMind's experimental cyclone model alongside physics-based models. Google says Weather Lab is a research tool, not an official warning service.[11][14]

    Sources

    Each numbered claim is a statement we checked against the sources listed with it. Status shows how well established it is.

    1. [1]

      Google DeepMind's GraphCast, described in Science in November 2023, gave more accurate predictions than ECMWF's HRES forecast on more than 90% of 1,380 test variables and lead times. confirmedas of 2023-11-14

    2. [2]

      GenCast, a probabilistic machine-learning weather model published in Nature in December 2024, had greater skill than ECMWF's ENS ensemble on 97.2% of 1,320 evaluated targets. confirmedas of 2024-12-04

    3. [3]

      Google DeepMind's WeatherNext 2, announced in November 2025, uses a Functional Generative Network and surpasses the previous WeatherNext model on 99.9% of variables and lead times from 0 to 15 days. confirmedas of 2025-11-17

    4. [4]

      GraphCast is a graph neural network trained on four decades of ECMWF's ERA5 reanalysis data, which is itself built from observations using traditional numerical weather prediction. confirmedas of 2023-11-14

    5. [5]

      GraphCast takes the weather state from six hours ago and the current state, predicts six hours ahead, and repeats the step to forecast up to 10 days. confirmedas of 2023-11-14

    6. [6]

      GraphCast makes a 10-day forecast in less than a minute on a single Google TPU v4 machine, while a conventional HRES forecast can take hours on a supercomputer with hundreds of machines. confirmedas of 2023-11-14

    7. [7]

      The GenCast paper notes that earlier ML weather models focused on single deterministic forecasts and remained less accurate and reliable than the best physics-based ensemble forecasts. confirmedas of 2024-12-04

    8. [8]

      GenCast was announced by Google DeepMind in December 2024 as a model delivering forecasts up to 15 days ahead. confirmedas of 2024-12-04

    9. [9]

      GenCast generates an ensemble of 15-day global forecasts at 0.25° resolution for more than 80 variables in 8 minutes. confirmedas of 2024-12-04

    10. [10]

      WeatherNext 2 can produce hundreds of possible weather outcomes from one starting point, each in under a minute on a single TPU. confirmedas of 2025-11-17

    11. [11]

      In June 2025 Google DeepMind and Google Research launched Weather Lab, with an experimental AI cyclone model that generates 50 scenarios up to 15 days ahead, and began sharing its predictions with the US National Hurricane Center. confirmedas of 2025-06-12

    12. [12]

      Google DeepMind said its internal testing found the experimental cyclone model's track and intensity predictions as accurate as, and often more accurate than, physics-based methods. confirmedas of 2025-06-12

    13. [13]

      Weather Lab displays live and historical cyclone predictions from Google's AI weather models alongside physics-based models from ECMWF. confirmedas of 2025-06-12

    14. [14]

      Google DeepMind says Weather Lab is a research tool and its live predictions are not official warnings. confirmedas of 2025-06-12

    15. [15]

      Google DeepMind's public WeatherNext repository contains code for WeatherNext 2 and the earlier GraphCast and GenCast models, and daily WeatherNext 2 forecast feeds are offered through Google Cloud, Weather Lab and Open-Meteo. confirmedas of 2026-10-10

    16. [16]

      ECMWF took its Artificial Intelligence Forecasting System (AIFS) into operations on 25 February 2025, running side by side with its physics-based IFS. confirmedas of 2025-02-25

    17. [17]

      On 17 December 2025 NOAA deployed operational AI-driven global models, AIGFS, AIGEFS and the hybrid HGEFS, which NOAA says deliver more accurate guidance using a fraction of the computing resources. confirmedas of 2025-12-17

    18. [18]

      ECMWF publishes the weights of AIFS Single v1.0, the first operationally supported AIFS model, under a CC BY 4.0 licence. confirmedas of 2026-10-10

    19. [19]

      ECMWF lists known AIFS limitations including blurred forecast fields at longer lead times, reduced skill in the stratosphere and reduced intensity of some high-impact systems such as tropical cyclones. confirmedas of 2026-10-10

    Revision history (2)
    1. Page created.
    2. Added Weather Lab's experimental cyclone model and National Hurricane Center partnership, plus public code and forecast feeds.

    Created Oct 10, 2026. Last reviewed by an editor on Oct 10, 2026. Next scheduled review: Jan 10, 2027.

    Cite this page

    "WeatherNext (GraphCast, GenCast)." ContentLora, updated Oct 10, 2026. https://contentlora.com/wiki/weathernext

    Spotted an error? Suggest a correction or emailcorrections@contentlora.com.

    Keep exploring

    Connections
    Google DeepMind Developed byECMWF AIFS Benchmarked against ECMWF
    ☰ Contents