We report MERLIN, VLA, OCRA-p, VLBA, Effelsberg and GMRT observations beginning 4.5 days after the discovery of RS Ophiuchi undergoing its 2006 recurrent nova outburst. Observations over the first 9 weeks are included, enabling us to follow spectral development throughout the three phases of the remnant development. We see dramatic brightening on days 4 to 7 at 6 GHz and an accompanying increase in other bands, particularly 1.46 GHz, consistent with transition from the initial free phase to the adiabatic expansion phase. This is complete by day 13 when the flux density at 5 GHz is apparently declining from an unexpectedly early maximum (compared with expectations from observations of the 1985 outburst). The flux density recovered to a second peak by approximately day 40, consistent with behaviour observed in 1985. At all times the spectral index is consistent with mixed non-thermal and thermal emission. The spectral indices are consistent with a non-thermal component at lower frequencies on all dates, and the spectral index changes show that the two components are clearly variable. The estimated extent of the emission at 22 GHz on day 59 is consistent with the extended east and west features seen at 1.7 GHz with the VLBA on day 63 being entirely non-thermal. We suggest a two-component model, consisting of a decelerating shell seen in mixed thermal and non-thermal emission plus faster bipolar ejecta generating the non-thermal emission, as seen in contemporaneous VLBA observations. Our estimated ejecta mass of 4+/-2x10^{-7} M_\odot is consistent with a WD mass of 1.4 M_\odot. It may be that this ejecta mass estimate is a lower limit, in which case a lower WD mass would be consistent with the data.