Gas is trapped in polar ice sheets at 50–120m below the surface and is therefore younger than the surrounding ice. Firn densification models are used to evaluate this ice age-gas age difference (1age) in the past. However, such models need to be validated by data, in particular for periods colder than present day on the East Antarctic plateau. Here we bring new constraints to test a firn densification model applied to the EPICA Dome C (EDC) site for the last 50 kyr, by linking the EDC ice core to the EPICA Dronning Maud Land (EDML) ice core, both in the ice phase (using volcanic horizons) and in the gas phase (using rapid methane variations). We also use the structured 10Be peak, occurring 41 kyr before present (BP) and due to the low geomagnetic field associated with the Laschamp event, to experimentally estimate the 1age during this event. Our results seem to reveal an overestimate of the 1age by the firn densification model during the last glacial period at EDC. Tests with different accumulation rates and temperature scenarios do not entirely resolve this discrepancy. Although the exact reasons for the 1age overestimate at the two EPICA sites remain unknown at this stage, we conclude that current densification model simulations have deficits under glacial climatic conditions. Whatever the cause of the 1age overestimate, our finding suggests that the phase relationship between CO2 and EDC temperature previously inferred for the start of the last deglaciation (lag of CO2 by 800±600 yr) seems to be overestimated.