The Japan Meteorological Agency (JMA) upgraded its supercomputing system in March 2006. The new system has about 28 times as fast as the previous system. With the new computing system, JMA plans to upgrade its global model from the current TL319/L40 resolution into the higher TL959L60 resolution in November 2007. Since we have been keeping the global model resolution (TL319/L40) in spite of the introduction of the new supercomputing system, we could carry out much more data assimilation experiments than the usual condition and made much progress in FY2006. Those are, the introduction of new satellite data (microwave imagers, AP-RARS data and GPS (CHAMP) radio occultation refractivity data) and variational bias correction method, and improvement on the way to use the ATOVS radiances and atmospheric motion vectors. In this talk, I will introduce these recent changes in the operational system briefly. Moreover, we have been testing the local ensemble transform Kalman filter (LETKF) technique using JMA global model at a TL159/L40 resolution, the same version as the one used in the JMA ensemble prediction system. The latest results showed that the anomaly correlations of 500 hPa geopotential height forecast in the northern hemisphere became comparable with the operational 4D-Var system, though it was not the case in the southern hemisphere. This talk also covers some of the recent developments of the LETKF.