EPT-2 Family

Jua's flagship state-of-the-art weather model, released in April 2025. EPT-2 outperforms leading public AI weather models including Microsoft Aurora, DeepMind's GraphCast, ECMWF's AIFS, and our previous EPT-1.5 model. For more information on our EPT-2 family of models, take a look at our EPT-2 Technical Report.

EPT-2 Early offers dissemination 2.5 hours earlier than EPT-2 to anticipate market movements.

EPT-2 RR is our rapid refresh forecast model, with our fastest dissemination times and 24 updates per day.

EPT-2e is our 30 member, perturbation-based ensemble model of EPT-2 for probabilistic forecasting. EPT-2e significantly surpasses the HRES ensemble mean—long considered the gold standard for medium- to long-range forecasting—while operating. Ensemble statistics are also available for EPT-2e through the platform, API and SDK.

Technical Specifications

Coordinates

Name	Description	Units
time	Initialization time of forecast at UTC	Absolute time (UTC)
prediction_timedelta	Lead time relative to initialization time	Timedelta
latitude	Latitudinal coordinate	Degrees true, range [90, -90)
longitude	Longitudinal coordinate	Degrees true, range [-180, 180)

Available Weather Parameters

EPT-2 uses standardized variable naming conventions. For full details on the naming structure, see Weather Variable Naming.

Parameter Code	API Name	Description	Units
2t	air_temperature_at_height_level_2m	Air temperature at 2m	Kelvin
2d	dew_point_temperature_at_height_level_2m	Dew point temperature at 2m	Kelvin
r	relative_humidity_at_height_level_2m	Relative humidity at 2m	%
10si	wind_speed_at_height_level_10m	Horizontal wind speed at 10m	m s⁻¹
10wdir	wind_direction_at_height_level_10m	Horizontal wind direction at 10m	Degrees true
100si	wind_speed_at_height_level_100m	Horizontal wind speed at 100m	m s⁻¹
100wdir	wind_direction_at_height_level_100m	Horizontal wind direction at 100m	Degrees true
msl	air_pressure_at_mean_sea_level	Mean Sea-level Pressure	Pa
lcc	low_type_cloud_area_fraction	Low Cloud Cover	%
mcc	medium_type_cloud_area_fraction	Medium Cloud Cove	%
hcc	high_type_cloud_area_fraction	High Cloud Cover	%
tp	precipitation_amount_sum_1h	Total Precipitation	mm m⁻²
ssrd	surface_downwelling_shortwave_flux_sum_1h	Surface solar downward irradiance	J m⁻²
fdir	surface_direct_downwelling_shortwave_flux_sum_1h	Direct solar radiation at the surface	J m⁻²
z_500	wind_direction_at_height_level_100m	Geopotential	m² s⁻²

Note: For information on the previous naming convention used in EPT-1.5, see Legacy Naming.

Additional Variables on request:

Parameter Code	API Name	Description	Units
tcwv	atmosphere_mass_content_of_water_vapor	Total column vertically-integrated water vapour	kg m⁻²
t	air_temperature_at_pressure_level	Temperature at a specified pressure level	K
u	eastward_wind_at_pressure_level	U component of wind	m s⁻¹
v	northward_wind_at_pressure_level	V component of wind	m s⁻¹
q	specific_humidity_at_pressure_level	The mass of water vapour per kilogram of moist air at a pressure level	kg
z	geopotential_at_pressure_level	The gravitational potential energy of a unit mass	m² s⁻²

• Pressure levels: 5000, 10000, 15000, 20000, 25000, 30000, 40000, 50000, 60000, 70000, 85000, 92500, 100000

Hindcasts Details

• Full hindcast dataset available for performance analysis and evaluation

• Hindcasts are global with a grid of 2160x4320

• Hindcasts are from 2023-01-01 to 2024-12-28

• Hindcasts are run 4 times a day at 00, 06, 12, 18

• Hindcasts are chunked as (time: 1, prediction_timedelta: 144, lat: 120, lon: 240)

  • Each chunk is 10 GB

• Forecasts are hourly for the first 10 days and 6 hourly from days 10 to 20

• Hindcasts are not available for EPT-2e

For migration information from EPT-1.5 to EPT-2, including parameter naming changes, see EPT-1.5 to EPT-2 Migration Guide.

Performance Benchmarks

EPT-2 consistently outperforms other leading weather models, including:

• Microsoft Aurora

• DeepMind's GraphCast

• ECMWF's AIFS

• Our previous EPT-1.5 model

Dissemination times

EPT-2 also features an Early version disseminated 2.5 hours earlier.

For detailed dissemination times of EPT-2 and EPT-2 Early, please refer to the Dissemination Times page.