An archive of ECMWF IFS operational analysis, every 6 hours from 2011-01-01 to 2026-09-30, on a N320 reduced Gaussian grid. It contains a subset of variables of the IFS model, serving as input for AI weather models such as ECMWF's AIFS. This is the analysis ECMWF used to start its own high-resolution forecasts. It is ECMWF's best estimate of the atmosphere every 6 hours, made by 4D-Var data assimilation. This dataset is an Icechunk repository on the N320 reduced Gaussian grid (~31 km, 542,080 points). It is regridded from the operational model resolution (O640 in 2011-2015, O1280 from 2016) to N320 using ECMWF MIR. The variables are the input set of data-driven weather models such as ECMWF's AIFS. That makes it ready for training, fine-tuning, initialising and verifying AI weather models without first building a GRIB pipeline.