Gyrochronology can yield useful ages for field main-sequence stars, a regime where other techniques are problematic. Typically, gyrochronology relations are calibrated using young (≲2 Gyr) clusters, but the constraints at older ages are scarce, making them potentially inaccurate and imprecise. In order to test the performance of existing relations, we construct samples of stellar pairs with coeval components, for a range of ages and with available rotation periods. These include randomly paired stars in clusters, and wide binaries in the Kepler field. We design indicators that, based on the measured rotation periods and expectations from gyrochronology, quantify the (dis)agreement between the coeval pairs and the gyrochronology calibrations under scrutiny. Our results show that wide binaries and cluster members are in better concordance with gyrochronology than samples of randomly paired field stars, confirming that the relations have predicting power. However, the agreement with the examined relations decreases for older stars, revealing a degradation of the examined relations with age, in agreement with recent works. This highlights the need for novel empirical constraints at older ages that may allow revised calibrations. Notably, using coeval stars to test gyrochronology poses the advantage of circumventing the need for age determinations while simultaneously exploiting larger samples at older ages. Our test is independent of any specific age-rotation relation, and it can be used to evaluate future spin-down models. In addition, taking gyrochronology at face value, we note that our results provide new empirical evidence that the components of field wide binaries are indeed coeval.