At a guess, the official range 100-25,600 could be fairly close to what ISO 12232 recommends as "ISO speed latitude", with 100 being about the well-saturation based minimum or "Ssat" value, and 25,600 being around the upper limit based on barely acceptable 10:1 SNR in the mid-tones, "Ssnr10". Basic photon counting says that beyond about 25,600, the mid-tone SNR level will be horribly low at full 16MP resolution, but down-sampling to about 1MP or less for publishing in the sports section of a newspaper or website could give quite usable results. Roughly, a 16-to-1 downsampling could increase the per-pixel Ssnr10 speed by a factor of 16; with low enough read noise from the sensor, SNR=10:1 just needs about 100 photons counted per pixel (not per photosite). In this context, note the new "small raw option".
ISO numbers are getting uncomfortably large, how about starting to use the good old logarithmic DIN again? 400 000 ISO would be 57 DIN, yes?
Yes, I like to think in terms of "stops above ISO 100", so the official range of 100-25,600 is like "0 to 8 stops" and the expanded range of 50-409,600 is "-1 to 12". The former German DIN standard is roughly this, but in one-third stop increments, counting from a bit below ISO speed of 1.
P. S. To combine two previous comments, the exposure index scales are ISO = ASA =100*2(DIN-21)/3 and "stops above ISO100" = (DIN-21)/3.