Context. Microlensing and reverberation mapping measurements of quasar accretion disk sizes and temperature gradients disagree with predictions from thin-disk theory. Previous microlensing results relied on heterogeneous wavelength coverage, primarily UV broad emission lines from small samples, which only probe the disk out to a typical radius of ≲5 light days on average. Aims. We used microlensing estimates from a homogeneous sample of near-infrared observations of lensed quasars (21 image pairs from seven lens systems) to extend the disk size measurements out to 14 light days. This analysis leveraged narrow emission lines, which provide a more reliable microlensing-free baseline than cores of broad emission lines. Methods. We derived Bayesian estimates of the accretion disk size that reproduce the observed microlensing magnifications, as simulated from magnification maps. Results. The sizes based on narrow emission lines yield a logarithmic slope of p = 0.68 ± 0.23, consistent with prior estimates corresponding to inner disk regions (r ≲ 5 light days). Conclusions. Using a new homogeneous near-infrared dataset that allowed us to reach radial distances of up to 14 light days, we found that the accretion disks in these previously unexplored regions are also larger and their temperature gradients are steeper than predicted by thin-disk theory. The increased precision allowed us to reject the theoretical logarithmic slope p = 4/3 at the 98% confidence level. Any hypothesis invoking contamination through the broad line region to explain this discrepancy must account for how this contamination modulates the underlying accretion disk such that the combination of the two results in a power law with logarithmic slope p = 0.68 ± 0.23 across a broad wavelength baseline spanning from ~X-Ray to ~5000 Å.