AFAIK the display one goes to 6 1/2 digits. More is of little use anyway because of noise an drift. The ADC itself gets close (but AFAIK just a little short) to 0.1 ppm resolution with 10 PLC integration, but there is quite some noise. So the last few bits are not stable. The limiting factor is not so much the theoretical resolution / quatization at the ADC but the noise and drift. With averaging over multiple conversions the noise goes down, but only relatively slow, like the square root. With 100 PLC = 10 x 10 PLC averaged the 7 th digit may get somewhat stable (at least near zero).
Like with many other meters the result send via GPIB can have more digits. The last ones are not at all meaningfull as only a limitited number of values are possible. With enough samples the same 9 digits numbers will repeat. The reading are not the raw ADC samples but the ADC samples multiplied with a high resolution (e.g. floating point, or maybe 32 bit) scale factor. If using the AZ mode there is the additional difference to the zero reading.
One can ignore the last digits, but it does not hurt if the computer takes them into acount (e.g. for averaging, calculating a slope or noise).