GraphCast
Graph-neural-network weather model making a 10-day global forecast in under a minute on one TPU v4, beating ECMWF HRES on most targets.
Learned simulator on a 0.25-degree grid, autoregressive on a multi-scale mesh. Blog reports better accuracy than HRES on more than 90% of 1,380 verification targets (99.7% in the troposphere). ECMWF began running it live. Code open-sourced.
- Date
- Tuesday, 14 November 2023
- Lab
- Google DeepMind
- Kind
- model
- Access
- research preview
Figures
| Measure | Value | Measured by |
|---|---|---|
| Targets beating ECMWF HRES | >90% of 1,380 99.7% of tropospheric targets | company |
Published in Science 2023-11-14. Code open-sourced; ECMWF ran it live. The paper's own abstract cites a different tally (89.3% of 2,760 variable/lead-time combinations), so figures differ by source.
Sources
- deepmind.google/blog/graphcast-ai-model-for-faster-and-more-accurate-global-weather-foreca
- www.technologyreview.com/2023/11/14/1083366/google-deepminds-weather-ai-can-forecast-extre
This record was checked against its sources on 6 October 2026. How we check