I can't really say if these devices are good or bad. However, their article bother me, as it does not actually tell you how this works.
While it is true that 32-bit float can store a huge dynamic range, the interesting bit is how you do the AD conversion to get those float values. If it were just a regular 32-bit ADC doing the conversion, it stands to reason that converting the output to floating point has no benefit over jut storing the 32-bit integer. If you can store very quite signals, but there is much more converter noise than signal, what is the benefit?
Now they don't tell us how they do it. Maybe there is some super novel technology here. More likely it is just a system of multiple AD conversions happening at the same time with different reference voltages. The signal actually stored is then picked from the converter with the best reference for the current input signal.
The problem with such systems is that they often have worse noise performance than systems where you manually set the gain. Suspiciously if you look at the specs of their F3 field recorder there is neither S/N-Ratio nor THD+N. And looking at the specs also uncovers something else that seems like a blatant lie: They claim you can record the loudest noises on earth, never clip, no worries about your input signal. Yet there is a maximum input level of +4dBu (Mic) and +24dBu (Line) specified. So, put a line signal into a mic input and you'll probably still clip, quite contrary to the marketing claim made.