It varies. There are many subtle aspects to consider:
- Crosstalk between open contacts, and separate sets
- Stubs, reflections and impedance mismatch along the length of the contacts
- Microvolt DC thermal errors (easily solved with latching relays)
- Contact bounce
- Arcing at higher voltages
- DC resistance; repeatability (how it varies between contact closure events); possibly rectifying contacts (due to oxidation)
- DC leakage (mostly over the surfaces of plastic insulators, but including corona discharge at HV)
Unless you're doing extremely sensitive or high current applications, we can safely ignore the contact-closed properties. And ditto, but for high voltages instead, we can ignore the contact-open properties. And for low power signals, we can ignore arcing as well.
That basically leaves frequency response. For a relay of that size (about 1cm), the contacts and pin lengths are on the order of 1cm as well, and so we expect to see quirks (humps and valleys in the s-parameters when open and closed) in the GHz range.
At low frequencies (say, under 1GHz), all those properties are asymptotic, and can be safely approximated as capacitances between open contacts, and inductances (including mutual inductances) between closed contacts.
Even a generic, signal type, not-RF relay has enviable properties: on the order of ~2pF per contact (to whichever nearby contacts, or the frame), and ~10nH per closed contact pair. The impedance is broadly around 100 ohms (as most things are!), and the "cutoff" frequency (not a -3dB point!) on the order of 1GHz.
The cutoff frequency is simply where behavior becomes more interesting, i.e., the lumped element model breaks down. The response ceases to be asymptotic around here, and you probably won't get useful switching behavior (i.e., on/off ratio).
For a relay used at very different impedances (i.e., very far above or below ~100 ohms), one or the other lumped equivalent parameter (L or C) will become significant, at a proportionally lower frequency (say, for a circuit impedance around 10kohm or 1 ohm, there will be a lowpass cutoff around 10MHz).
The closer you get to the true characteristic impedance of the device, the more bandwidth you'll be able to use, up to its full frequency range (~1GHz). This is one possible way of measuring its characteristic impedance. (But, because a generic type relay won't have an impedance-controlled design, it won't have a stable characteristic at higher frequencies, so there isn't much value in measuring it precisely. An RF relay will be designed this way, however, and will be able to operate well above its LF 'cutoff' frequency.)
Tim