Skip to content
ContentLora

    Tip: press / anywhere to search.

    product

    ECMWF AIFS (Artificial Intelligence Forecasting System)

    Also known as AIFS, AIFS Single, AIFS ENS, Artificial Intelligence Forecasting System

    The Artificial Intelligence Forecasting System (AIFS) is the European Centre for Medium-Range Weather Forecasts' machine-learning model, made operational on 25 February 2025 alongside its physics-based IFS.[1] ECMWF says it improves tropical cyclone tracks by up to 20% while using about 1,000 times less energy per forecast.[2]

    Editor reviewedUpdated AI for scienceArtificial intelligenceScience
    Key facts

    The AIFS marks the point when AI weather forecasting moved from tech-company research into a public forecasting centre’s daily operations. ECMWF, whose physics-based forecasts were the benchmark early AI models were measured against, put its own machine-learning model into service on 25 February 2025.[1][3]

    What it is

    The AIFS is a machine-learning forecasting system that runs side by side with ECMWF’s Integrated Forecasting System (IFS), the centre’s traditional physics-based model.[1] A deterministic version came first; the ensemble version, which produces many forecasts to express uncertainty, became operational on 1 July 2025.[4]

    How it works

    ECMWF describes the AIFS as a graph neural network encoder and decoder wrapped around a sliding-window transformer processor. It is trained on ERA5 reanalysis and on ECMWF’s own operational analyses, the best estimates of the current state of the atmosphere.[5] It runs four times a day alongside the physics-based model, and its forecasts are public under ECMWF’s open data policy.[6]

    ECMWF has also opened the model itself. The weights of AIFS Single v1.0, the first operationally supported AIFS model, are published under a CC BY 4.0 licence, so researchers and companies can run and adapt it.[7] Google DeepMind likewise publishes code for its weathernext models, with forecast feeds offered through Google Cloud and other channels.[8]

    Performance

    ECMWF reports that the AIFS outperforms state-of-the-art physics-based models on many measures, including tropical cyclone tracks with gains of up to 20%.[2] It also reports about a 1,000-fold cut in the energy needed to make a forecast.[2] These are ECMWF’s own figures; the centre continues to run the IFS in parallel.[1]

    ECMWF is also explicit about weaknesses. Its model card lists blurring of forecast fields at longer lead times, reduced skill in the stratosphere and reduced intensity of some high-impact systems such as tropical cyclones.[9] In other words, the AIFS can place a cyclone’s track well while underplaying how strong it is.[2][9]

    Who uses it

    ECMWF highlights the renewable-energy sector, where the AIFS helps forecast surface solar radiation and wind speeds at turbine height.[10]

    Context

    The AIFS arrived after tech-company models showed what ML could do. Google DeepMind’s GraphCast beat ECMWF’s high-resolution HRES forecast on more than 90% of test targets in 2023, and GenCast beat ECMWF’s ENS ensemble on 97.2% of targets in 2024.[3][11] Those models were trained on ECMWF’s own ERA5 reanalysis.[12] ECMWF’s physics-based forecasts remain a common yardstick: Google’s Weather Lab shows its AI cyclone predictions next to ECMWF’s physics-based models.[13] In the United States, NOAA deployed its own AI-driven global models, including a hybrid AI-physics ensemble, in December 2025.[14][15]

    Open questions

    Running AI and physics models in parallel lets forecasters compare them day by day.[1] Because AI models learn from past weather, their developers expect to keep retraining them: Google DeepMind wrote that GraphCast will evolve as weather patterns change in a warming climate and better data become available.[16] See weathernext for the Google DeepMind models.

    Questions readers ask

    Has ECMWF replaced its physics model with AI?

    No. The AIFS runs side by side with the physics-based IFS.[1]

    Is there an ensemble version?

    Yes. ECMWF made the ensemble AIFS operational on 1 July 2025.[4]

    Can I download the AIFS?

    Yes. ECMWF publishes AIFS Single v1.0 weights under a CC BY 4.0 licence, and AIFS forecasts are available under its open data policy.[7][6]

    What are its known weaknesses?

    ECMWF lists blurring at longer lead times, weaker skill in the stratosphere and reduced intensity for some high-impact systems such as tropical cyclones.[9]

    Who benefits from AIFS forecasts?

    ECMWF highlights renewable energy, for example forecasts of solar radiation and turbine-height wind speeds.[10]

    Sources

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

    1. [1]

      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

    2. [2]

      ECMWF says the AIFS outperforms physics-based models on many measures, including tropical cyclone tracks with gains of up to 20%, and uses about 1,000 times less energy per forecast. confirmedas of 2025-02-25

    3. [3]

      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

    4. [4]

      ECMWF took the ensemble version of the AIFS into operations on 1 July 2025. confirmedas of 2025-07-01

    5. [5]

      ECMWF describes the AIFS as a graph neural network encoder and decoder with a sliding-window transformer processor, trained on ERA5 reanalysis and ECMWF's operational analyses. confirmedas of 2024-08-07

    6. [6]

      The AIFS runs four times a day alongside ECMWF's physics-based model, and its forecasts are public under ECMWF's open data policy. confirmedas of 2026-10-10

    7. [7]

      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

    8. [8]

      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

    9. [9]

      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

    10. [10]

      ECMWF says the AIFS will help renewable-energy users predict quantities such as surface solar radiation and wind speeds at turbine height. confirmedas of 2025-02-25

    11. [11]

      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

    12. [12]

      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

    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]

      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

    15. [15]

      NOAA's AI models stem from Project EAGLE, a multi-year collaboration across NOAA research, the National Weather Service, academia and industry, and combine AI and physics-based approaches. confirmedas of 2025-12-17

    16. [16]

      Google DeepMind wrote that GraphCast will need to evolve and improve as weather patterns change in a changing climate and higher-quality data become available. confirmedas of 2023-11-14

    Revision history (2)
    1. Page created.
    2. Added the AIFS architecture, open data and open weights, and ECMWF's list of known limitations.

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

    Cite this page

    "ECMWF AIFS (Artificial Intelligence Forecasting System)." ContentLora, updated Oct 10, 2026. https://contentlora.com/wiki/ecmwf-aifs

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

    Keep exploring

    Connections
    WeatherNext ECMWF forecasts were the …
    ☰ Contents