A few highlights

Python 2 is no longer supported; MNE-Python now requires Python 3.5+
New tutorials and examples on sleep stage classification, data simulation using subject's anatomy, how to use EEG montages on fsaverage
New module to simulate SourceEstimates
Improved performance using CUDA, better copy management, better dispatching of the computation over the channels
New fetchers for polysomnography (PSG) recordings from Physionet and fsaverage template
Better support for source reconstruction with beamformers and other inverse models
Improved UI and visualizations for topomaps, Raw and Epochs objects
Better support for Annotations
Support to compute power envelope correlations on brain parcellation for rest data
Better rendering of the coregistration
Added partial support for PyVista as a 3D backend that can replace mayavi
Added support for Raw, Epochs and Evoked noise simulation
Added new parcellation (448-labels subdivided aparc) and improve support including morphing of the labels
Better support for the TFR objects

And we caught and fixed more than 50 bugs!

Notable API changes

Deprecation of mne.realtime module to become entire project in the MNE echosystem
Deprecation of mne.io.find_edf_events, raw.estimate_rank
Reading BDF and GDF files with mne.io.read_raw_edf is deprecated and replaced by mne.io.read_raw_bdf and mne.io.read_raw_gdf
The signatures of mne.preprocessing.ICA, mne.simulation.add_noise and mne.simulation.add_chpi have changed
Added overwrite parameter in mne.Epochs.save
mne.minimum_norm.apply_inverse now returns mne.VolVectorSourceEstimate when needed
Annotations are now kept sorted by onset
peak_finder should be imported as: from mne.preprocessing import peak_finder