Since its inception, nDAX has used the default DAX audio mode, which is 32-bit floating point samples at 48kHz. This uses a bandwidth of about 1.64Mbit/s.
But it seems that Flex's own FlexDAX exclusively uses the "reduced bandwidth" DAX mode, which is 16-bit integer samples at 24kHz, which uses a bandwidth of about 468kbit/s, for a savings of 70%.
Since all Flex waveforms run at 24kHz, there should be no loss from the lower sample rate, and the loss of dynamic range should be minimal (the ADC is 16 bits to begin with).
Besides saving bandwidth (improving performance on weak wifi, cellular links, etc.), the low-bandwidth mode will probably also reduce CPU usage in most scenarios, since most applications want 8-bit or 16-bit data anyway, and converting from floating-point comes at a cost.
If you really need the high-bandwidth mode (a valid case might be receiving very weak signals with AGC OFF), it can be re-enabled with the -high-bw option.