HP5061A is a classic. Spec says that it is +/- 1 x10^-11 relative to primary standard for averaging times > 1 second. Weighing in at 60lbs, I don’t think you’ll be taking it out portable, though NPL used to have to drive one over to Rugby to check MSF!
For GPS, I have never worked through all of the correction procedures to find out how close to primary standard can be managed, but I recall that around +/- 1 x 10^-13 is achievable, though not with the system we have at 1 second averaging time.
We have a work project that is looking at comparing primary standards using optical fibre links, but that has its own challenges.
This is a psychologically dangerous area to get interested in- see the Time Nuts web-site.
Alwyn Seeds, Director
114 Beaufort Street,
SynOptika Ltd., Registered in England and Wales: No. 04606737
Registered Office: 114 Beaufort Street, London, SW3 6BU, United Kingdom.