Given your requirements and making quite a few assumptions the HL series is your best bet.
The HL6 is the quad band 2G variant. Depending on network conditions (which are controlled by the network and not the module) it is possible to get it down to 1mA (potentially marginally lower), while attached to the network and with a session running no data being sent), this is about as good as you can get. Peak data for the actual transmission is as usual high for 2G but it is optimised as much as possible.
The G variant of the HL6 also has a SirF stat V chipset on it as well.
The firmware for HL8 is still under development and current consumption data will not be accurate until it is officially commercially available next year. What I can say is that 2G figures are going to be worst case (modulation scheme is not optimised for current consumption). The figures that have been posted will be accurate and measured values against a test set
The primary parameter that affects this will be how well it can see the base station, all base stations like to be able to see the units they are communicating with at -95dBm, if you are 25Km away or in a basement then the base station will command the unit to transmit at a high power to et to this level hence higher power consumption.
When in idle the consumption is primarily determined by two things.
How well the hardware has been designed around the module i.e. if you have incorrect voltage levels on GPIO’s then there could be significiant leakage current.
The level of DRX (discontinuous reception) the network commands the unit to apply which means it should listen to 1 in x slots where x can be 1-9. Obviously the less it needs to power the receiver up the lower the current consumption.
As Andy has pointed out the lowest consumption will be off, in this case all of the modems will be consuming in the order to 10’s of uA’s to run the RTC and charge whatever backup cell there is connected to it. However if you do not need the RTC to run then you can just cut power to the Vbatt lines (would suggest this is god to implement as the ultimate reset in any design anyway).
Just to be clear about the use case. If the unit is in sleep mode (powered, registered onto the network but no external interfaces powered) and it is commanded to wake up (say through toggling DTR) to full power mode this will typically take 1-2 seconds.