Arctic summer sea ice has decreased dramatically over the last few decades, with a substantial part of this decline attributed to internal variability. However, models show large differences in their simulated internal variability, increasing projection uncertainty and complications with model-observation comparisons. Here we will present results that aim to quantify the contribution of internal variability in different models which provide large ensemble simulations, and compare them with estimates from observations. In particular, we are comparing five models from the CLIVAR multi-model large ensemble (CanESM2, CESM1, CSIRO MK36, GFDL ESM2M, and MPI ESM1) with observations. So far, we have found a large range in simulated pan-Arctic sea ice area standard deviation from 0.35 million km2 (CSIRO MK36) to 0.74 million km2 (CESM1) for mean September areas between 4.00-4.25 million km2. Spatially, the detrended standard deviation in the central Arctic is consistently over-represented in models compared to observations. Conversely, the marginal seas are simulated to have slightly below to several times below observed detrended standard deviation. Further analysis on a more regional scale will be done over the coming months to further characterize the realism of simulated internal variability in Arctic sea ice.