Version 1.13.0 (2026-09-09)

Dependencies

• Updated minimum for:

  • Core dependency matplotlib >= 3.9

  Changes implemented via CI action created by Thomas Binns. (#13899)

• Updated minimum for:

  • Optional dependency scikit-learn >= 1.5

  Changes implemented via CI action created by Thomas Binns. (#13916)

• Switch the default Qt bindings from PyQt6 to PySide6. As a result, mne[full] now includes PySide6 instead of PyQt6. Users who prefer PyQt6 can continue to use mne[full-pyqt6], and mne[full-pyside6] remains available as an alias for mne[full]. Changes by Clemens Brunner. (#13960)

• PyQt5 is no longer an officially supported Qt binding, by Eric Larson. (#13981)

• Updated minimum for:

  • Core dependency numpy >= 2.0, < 3

  • Core dependency scipy >= 1.14

  Changes implemented via CI action created by Thomas Binns. (#14023)

• Updated minimum for:

  • Optional dependency pyvista >= 0.44

  Changes implemented via CI action created by Thomas Binns. (#14048)

• Raised minimum versions of some optional dependencies to comply with MNE-Python’s SPEC0-like policy, and added them to the set kept up to date automatically by the spec_zero CI job, by Eric Larson:

  • Optional dependency dipy >= 1.9

  • Optional dependency mne-qt-browser >= 0.6

  • Optional dependency nibabel >= 5.2

  • Optional dependency nilearn >= 0.10

  • Optional dependency numba >= 0.60 (#14108)

• Updated minimum Python version to 3.11, by Thomas Binns. (#14140)

• Updated minimum for:

  • Core dependency numpy >= 2.1, < 3

  Changes implemented via CI action created by Thomas Binns. (#14193)

Bugfixes

• Fix bug where mne.io.read_raw_eyelink() fails to handle data starting with empty blocks or blocks starting with missing values, by new contributor Varun Kasyap Pentamaraju and Qian Chu. (#13571)

• Fix mode parameter in get_point_spread() and get_cross_talk() to correctly map public mode names ('max', 'svd') to internal names, expose previously hidden modes 'sum' and 'maxval' in the public API, and raise ValueError for invalid mode values, by new contributor Famous Raj Bhat. (#13754)

• Fixed incorrect label orientation for nodes in the 0–90° polar range (the 12–3 o’clock quadrant) of the connectivity circle plot, by new contributor Pavel Popov. (#13855)

• Fixed an indexing bug in fNIRS support in mne.io.BaseRaw.interpolate_bads() (and related methods) that could errantly use incorrect donor channels, by new contributor Kalle Makela. (#13856)

• Fix handling of optional dependencies that don’t specify a version attribute, by new contributor Baris Talar. (#13870)

• Improved error message when FreeSurfer executable is not found, by new contributor Ayushi Satodiya. (#13874)

• Fix mne.set_eeg_reference() so that passing a list of channel types with projection=False and ref_channels="average" now references each channel type independently by default (matching the projection=True path), by new contributor Ben Tang. Previously a single average was computed across the union of the listed types. Pass joint=True for the previous behavior. (#13920)

• Fix mne.preprocessing.eyetracking.interpolate_blinks() so that all annotation descriptions passed via match are interpolated over, rather than only BAD_blink ones, by new contributor Vincent Gao. (#13940)

• Fix bug in mne.read_annotations() where reading a .txt file containing a single channel-specific annotation raised an AttributeError, by new contributor Piotr Durka. (#13962)

• Fix bug where mne.preprocessing.write_fine_calibration() wrote internal MEG channel names (e.g. MEG0113) instead of the bare channel numbers (e.g. 113) that MaxFilter expects, breaking round trips of Neuromag calibration files, by new contributor Apoorva Verma. (#13999)

• Fix mne.preprocessing.eyetracking.read_eyelink_calibration() to decode .asc files as UTF-8 instead of ASCII, so non-ASCII characters in MSG lines no longer raise a UnicodeDecodeError, by new contributor Cedric Conday. (#14002)

• Fix the tube method of the abstract 3D renderer, whose signature was missing the opacity parameter that the PyVista implementation accepts, and whose docstring documented backface_culling, which tube does not take, while omitting normalized_colormap, which it does, by new contributor Anton Karpov. (#14125)

• Use a fixed normalized Kuiper-index cutoff of 0.3 for automatic CTPS ECG detection in mne.preprocessing.ICA.find_bads_ecg(), independent of sampling frequency and trial count, by new contributor Amirali Moradniaei. (#14279)

• Fix incorrect TFCE scores in mne.stats.permutation_cluster_test(), mne.stats.permutation_cluster_1samp_test(), and related functions when using 1D data or non-unit threshold steps. The cluster extent was always counted as 1 for 1D data, and the per-threshold height contribution did not match the reference values, by new contributor Antonio Elias Sánchez Soto. (#14286)

• Fixed a bug in mne.stats.spatio_temporal_cluster_1samp_test() and related spatio-temporal permutation cluster functions where, in rare cases, clusters that should have been joined by a chain of spatial and temporal adjacency steps were incorrectly reported as separate clusters, by Sharif Haason. (#13731)

• Make channel name labels easier to read in mne.viz.plot_alignment(), by Aman Srivastava. (#13773)

• Improved colocated OPM topomap handling by recognizing additional OPM coil types (FieldLine and Kernel), by Pragnya Khandelwal. (#13825)

• Fix bug where logging messages were sometimes errantly printed, by Eric Larson. (#13843)

• Add a memmap keyword argument to mne.io.BaseRaw.load_data(), allowing loading deferred raw data into a memory-mapped file after object creation, by Aniket Singh Yadav. (#13858)

• Completed triaxial OPM topomap grouping by rendering separate radial and tangential maps in plot_evoked_topomap() and plot_components(), and fixed shape-mismatch crashes in plot_projs_topomap() and animate_topomap() with colocated OPM channels, by Pragnya Khandelwal. (#13866)

• Fix bug with mne.preprocessing.ICA.plot_properties() when using reject in mne.preprocessing.ICA.fit(), by Eric Larson. (#13885)

• Fix bug where picks was ignored in mne.Epochs.apply_function() when channel_wise=False, by Thomas Binns. (#13894)

• Fix iteration errors in mne.io.read_raw_nihon() when reading log files, by Tom Ma. (#13915)

• Fix bug with mne.set_eeg_reference() not working when passing a single channel name as string, by Michael Straube. (#13917)

• Use mffpy-backed parsing for EGI MFF event tracks while tolerating nanosecond timestamps that some files store in beginTime, by Pragnya Khandelwal. (#13932)

• Fix channel ordering bug in mne.viz.plot_compare_evokeds() when axes="topo", by Lifeng Qiu Lin. (#13943)

• Fixed bug where mne.sys_info() would silently change the active Matplotlib backend (e.g. when ipympl is installed) as an import-time side effect, by Stefan Appelhoff. (#13963)

• Fix the error message raised when a requested FIR filter_length is too short, which reported half the actual transition bandwidth, by Cedric Conday. (#14005)

• Fix a crash (AttributeError on NoneType) when picking channels on a Raw whose _orig_units is None, by Cedric Conday. (#14006)

• Fixed several error messages that displayed literal {...} placeholder text instead of the intended values, due to a missing f string prefix (in mne.SourceEstimate, mne.io.read_raw_hitachi(), mne.preprocessing.compute_current_source_density(), and mne.preprocessing.maxwell_filter()), by Cedric Conday. (#14008)

• Fix spatial_colors is not "auto" as described by docstring in mne.viz.plot_evoked(). Fix empty topomap created when clicking after time range selection in mne.viz.plot_evoked(), by Lifeng Qiu Lin. (#14025)

• Fix bug where mne.concatenate_epochs() could silently concatenate epochs with conflicting event_id dicts, by Eric Larson. (#14051)

• Fixed two bugs in how SSS information is stored and read: mne.preprocessing.maxwell_filter() recorded which multipole moments it had kept in the wrong order (the filtered data itself was unaffected), and mne.compute_rank() reported the full MEG rank rather than the reduced SSS rank for data processed with MaxFilter™-3.0, by Eric Larson. (#14059)

• Fix figures not being displayed in notebook-style frontends when using a Matplotlib inline backend, by Mingjian He. (#14076)

• Fix Qt browser close messages from mne.io.Raw.plot() appearing in the wrong Jupyter notebook cell, by Mingjian He. (#14077)

• When concatenating Raw objects containing annotations, also merge the extras fields, by Marijn van Vliet (#14084)

• Fixed plot_evoked_topomap() and plot_sensors() (and the plot_sensors methods) to follow Matplotlib’s blocking behavior when show=True instead of returning immediately; plot_sensors keeps its block argument (default now None), by Clemens Brunner. (#14098)

• Fixed bug in mne.io.Raw.plot() where, with the matplotlib backend and group_by="selection", the channel-selection radio buttons no longer all appeared selected while in butterfly mode, by Eric Larson. (#14108)

• Fixed bug where dialogs of mne.io.Raw.plot() did not respect dark mode, by Eric Larson. (#14109)

• mne.chpi.filter_chpi() now smoothly interpolates the fitted cHPI/line amplitudes by default (interp="hann"), by Eric Larson. (#14112)

• Speed up mne.filter.notch_filter() and related methods when method="spectrum_fit" by caching repeated multitaper window setup, by Clemens Brunner. (#14116) (#14114)

• Fix bug in mne.gui.coregistration() where “Save Scaled Anatomy” failed with an Error scaling ... message and no scaled subject was created, by Eric Larson. (#14132)

• Fix bug in mne.preprocessing.find_bad_channels_maxwell() and mne.preprocessing.maxwell_filter(): data is processed in chunks, but when head_pos was provided, the head positions from the beginning of the recording were used for each chunk rather than from the chunk’s time window. In find_bad_channels_maxwell this affected every interval after the first, and in maxwell_filter every segment following a segment skipped due to skip_by_annotation, as well as every chunk containing no head position update at all, by Christian Brodbeck. (#14142)

• Allow mne.read_epochs_eeglab() to read epoched .set files without event information, by Daria Agafonova. (#14163)

• Add CUDA support to mne.io.Raw.apply_hilbert() and related data classes, allowing mne.preprocessing.annotate_muscle_zscore() to use n_jobs='cuda', by Daria Agafonova. (#14164)

• Fixed a bug where block=True in plotting functions such as mne.io.Raw.plot() ran the Qt application’s own event loop, which deadlocked or silently did not block when something else already owned that loop (for example inside a notebook kernel), and could not be interrupted, by Eric Larson. (#14179)

• Fix a bug that caused mne.io.Raw.pick() and mne.io.Raw.drop_channels() to fail on the result of mne.preprocessing.maxwell_filter() when movement compensation was enabled with head_pos, by Christian Brodbeck. (#14180)

• Fix bug where assigning into mne.Annotations.description (e.g., annotations.description[0] = "a_longer_description") silently truncated the new value, by storing descriptions using NumPy’s variable-width string dtype (numpy.dtypes.StringDType) instead of a fixed-width "<U" dtype, by Eric Larson. (#14186)

• Fixed a bug where volumetric source estimates plotted with signed data and the default MIP mapping (volume_options=dict(blending="mip")) rendered the negative half on top of the positive half from every viewing angle, by Eric Larson. (#14187)

• Fix mne.concatenate_epochs() failing on epochs cropped after baseline correction, by Kalle Makela. (#14188)

• Fix reading CTF datasets whose .hc file spells the head coils Nasion/LPA/RPA, by Seyed Yahya Shirazi. (#14191)

• Fix bug where clicking on the brain in mne.viz.Brain did not work with recent PyVista versions, by Eric Larson. (#14202)

• Preserve caller-owned memory-mapped preload files after Raw objects are destroyed, by Bruno Aristimunha. (#14213)

• Fix bug where mne.io.read_raw_ant() and mne.io.read_raw_curry() ignored projections and non-boolean preload values respectively when reading data lazily, by Bruno Aristimunha. (#14216)

• Fix bug where mne.io.anonymize_info() could set an incorrect date in info["proc_history"], causing an OverflowError when writing FIF files anonymized with a large daysback, by Eric Larson. (#14248)

• Fix bug where selecting epochs (e.g., epochs[0]) copied the entire data array, by Eric Larson. (#14262)

• Fix bug where mne.io.Raw.interpolate_to() and related methods with method="spline" did not center the target sensor positions on the fitted sphere origin, by Eric Larson. (#14266)

API changes by deprecation

• Rename standard_1005 and standard_1020 montages to colin27_1005 and colin27_1020, respectively. If you used standard_1005 or standard_1020, either switch to the new names colin27_1005 and colin27_1020 or use the new spherical montages spherical_1005, spherical_1010, or spherical_1020, which provide idealized template locations on a sphere (useful for plotting). By Clemens Brunner. (#13903)

• In mne.Evoked.animate_topomap() the vmin and vmax keyword arguments have been deprecated in favor of vlim and will be removed in MNE 1.14, by Eric Larson. (#13905)

• In mne.viz.iter_topography(), the on_pick callable for sub-figures now has a new parameter orig_fig that refers to the main figure. To preserve backward compatibility, it is made optional. By Lifeng Qiu Lin. (#13968)

• The block parameter of mne.time_frequency.Spectrum.plot_topo(), the legacy plot_psd_topo methods, and mne.viz.plot_raw_psd_topo is deprecated and will be removed in MNE 1.14; these plots now follow Matplotlib’s blocking behavior, by Clemens Brunner. (#14098)

• Deprecate the epochs parameter of mne.compute_covariance() in favor of inst; it will be removed in MNE 1.14, by Hamza Abdelhedi. (#14129)

• The matplotlib 3D backend for plotting source estimates (backend="matplotlib" in mne.viz.plot_source_estimates(), along with the spacing argument that only affects it) has been deprecated and will be removed in MNE 1.14; use a proper 3D backend such as pyvistaqt instead, by Eric Larson. (#14160)

• Add rng parameters to APIs that use randomness, while retaining keyword-only seed and random_state for compatibility (#9233 by Bruno Aristimunha). (#14199)

New features

• Speed up mne.stats.spatio_temporal_cluster_1samp_test() and related permutation cluster functions via precomputed sum-of-squares for sign-flip t-tests and SciPy connected-components clustering (~5x), by new contributor Sharif Haason. (#13731)

• Fix mne.preprocessing.nirs.beer_lambert_law() to support explicitly provided sd_distances during Beer-Lambert conversion. When valid source-detector distances are already present in raw.info, overriding them now emits a warning, and when all inferred distances are invalid and no explicit distances are provided, a ValueError is raised instead of silently producing zero concentrations, by new contributor Kalle Makela. (#13875)

• Added basic support for TD fNIRS data, by new contributor Zahra M. Aghajan, new contributor Julien Dubois, new contributor John Griffiths, Robert Luke, and Eric Larson. (#13938)

• Added public mne.viz.LayeredMesh class (previously private _LayeredMesh) with a smooth_mat attribute for efficient real-time source-space data streaming. Added Brain.layered_meshes attribute to access meshes after plotting, and a key parameter to mne.viz.Brain.add_data() for named overlays, by new contributor Payam Sadeghi-Shabestari. (#13970)

• Allow subclasses of MNEBrowseFigure to be passed to plot_raw/plot_epochs, as well as the corresponding plot(...) methods of the raw and epochs classes, by new contributor Frankie Robertson (#13979)

• Added meanlogratio baselining mode to mne.time_frequency.EpochsTFR.apply_baseline() and related functions, by new contributor Virginie van Wassenhove. (#14205)

• Added Generalizing across location (GAL) decoding and TimeGAL, a tutorial on Generalization Across Location (GAL) and Time-GAL decoding of EEG data, by Bruno Aristimunha and new contributor Alejandro Santos Mayo. (#14211)

• Speed up the permutation cluster tests (e.g., mne.stats.spatio_temporal_cluster_1samp_test()), especially with TFCE, by grouping cluster indices with a single sort instead of one pass over the data per cluster, by new contributor Luca Kaemmer. (#14217)

• Added 4 new ERP measures: peak latency, peak amplitude, fractional peak latency, and fractional area latency. By Jacob Woessner and new contributor Beliz Sertcan Gökmen. (#14221)

• Add mne.label_adjacency() and mne.volume_label_adjacency() for determining whether labels are adjacent (touching) or not, by new contributor Raphaël Bordas and Marijn van Vliet (#14226)

• Add mne.io.BaseRaw.get_annotation_spans() to convert annotation spans to the time reference used by Raw data and plotting methods, by new contributor Tim Anderson. (#14240)

• Add single-model JAMICA support to mne.preprocessing.ICA via method="jamica", by new contributor Sina Esmaeili. (#14247)

• Add mne.preprocessing.eyetracking.find_blinks() to detect blinks in eye-tracking data from pupil dropouts and return them as Annotations, by Stefan Appelhoff. (#12946)

• Add a GUI for interactive guided dipole fitting (mne.gui.dipolefit()), by Marijn van Vliet (#13074)

• It is now possible to concatenate raw objects with mne.concatenate_raws() as long as they inherit from BaseRaw, even if their specific types differ (e.g., Raw and RawArray), by Clemens Brunner. (#13263)

• Add a preprocessing example showing how to explore epoch quality before rejection using robust statistics (peak-to-peak amplitude, variance, and kurtosis) inspired by FASTER (Nolan et al., 2010) and Delorme et al. (2007), by Aman Srivastava. (#13710)

• Add annotation_colors parameter to mne.io.Raw.plot() and mne.Epochs.plot() to allow users to specify custom colors for annotations by passing a dict mapping annotation description strings to colors (for example, annotation_colors=dict(bad_segment="orange")), by Clemens Brunner. (#13838)

• Add dark theme support (theme="light", "dark", or "auto") to the 'matplotlib' browser backend used by mne.io.Raw.plot(), mne.Epochs.plot(), and mne.preprocessing.ICA.plot_sources(), by Clemens Brunner. (#13861)

• Add overrides parameter to mne.io.read_raw_brainvision() for reading non-spec-compliant .vhdr files where the header contradicts the actual layout (e.g. renamed BIDS siblings, missing MarkerFile=, truncated [Channel Infos]). Accepts a dict with keys data_fname, marker_fname, n_channels, sfreq, ch_names, units_fallback, data_orientation, data_format, and binary_format; see the function docstring for details. Missing or empty MarkerFile= is now also tolerated natively (per the BV Core Data Format spec). When MarkerFile= names a path that does not exist (common after BIDS renames), mne.io.read_raw_brainvision() now falls back to the co-located <stem>.vmrk next to the .vhdr and warns, instead of raising FileNotFoundError, by Bruno Aristimunha. (#13884)

• Add idealized spherical 10–05, 10–10, and 10–20 montages (spherical_1005, spherical_1010, spherical_1020), by Clemens Brunner. (#13903)

• Improved mne.Report by:

  • Using "agg" backend for 2D plots to avoid windows unnecessarily popping up

  • Improving add_stc() by using direct capture of brain images, allowing direct control of image size using stc_plot_kwargs["size"]

  • Improving add_evokeds() evoked topomap plotting by using mne.Evoked.animate_topomap() with butterfly=True rather than re-plotting each topomap in a loop

  Also improved mne.Evoked.animate_topomap() by deduplicating code with mne.Evoked.plot_topomap(). Changes by Eric Larson. (#13905)

• Add support for reading cross-talk and fine-calibration embedded in FIF files during acquisition into raw.info["cross_talk"] and raw.info["fine_calibration"], respectively, by Eric Larson. (#13911)

• Add support for interactive label browsing using hover=True in mne.viz.Brain.add_annotation(), by Eric Larson. (#13931)

• Added 'fsaverage_1005' montage, which is a slightly improved version of the existing 'colin27_1005' montage (known in 1.12 and older as 'standard_1005'). The electrodes have already been mapped to the fsaverage head surface. Changes by Eric Larson. (#13936)

• Add support for automatic cross_talk and fine_calibration adjustments in mne.preprocessing.maxwell_filter() when these are stored in mne.Info during acquisition (e.g., by newer MEGIN acquisition software), by Eric Larson. (#13942)

• Added the n_components parameter to ICA plotting and report generation functions to allow dynamic control over the number of components displayed by Athish M. (#13958)

• Added a volume_options["interpolation"] option to mne.viz.plot_source_estimates() and mne.viz.Brain.add_data() to control the interpolation method used for volume rendering. It can be "linear" (default) or "nearest". Changes by Eric Larson. (#13966)

• Include label and description fields when reading EGI mff files, by Martin Oberg. (#13967)

• Add TimeChange behaviour for mne.viz.plot_evoked_topo(), where time is marked with solid vertical line after clicking on subfigure axis. Add recursive parameter to mne.viz.ui_events.link() that enables linking to a target figure, as well as all figures that the target figure links to, by Lifeng Qiu Lin. (#13968)

• Add mask_label_params parameter to plot_topomap() and plot_evoked_topomap() to allow for customisation of masked channel labels, by Carina Forster. (#13969)

• Skip replotting of annotations on the horizontal scrollbar in the matplotlib browser backend, resulting in a significant speedup for traces with a large number of annotations by Frankie Robertson. (#13972)

• Add mouse wheel scrolling support to the 'matplotlib' browser backend to navigate through channels, by Clemens Brunner. (#13988)

• add_data() now supports multiple simultaneous overlays via remove_existing=False and distinct key values. The remove_existing parameter now accepts True / False in addition to None. By Payam Sadeghi-Shabestari. (#13995)

• Add on_outside parameter to mne.Epochs and mne.EpochsArray classes to control the behaviour when events exist outside the data range. The options are to 'warn' (new default), 'raise', or 'ignore' (old behaviour), by Cedric Conday. (#14004)

• mne.viz.Brain’s silhouette option now accepts a spacing string (e.g. "ico5", now the default) to decimate using the same vertex-picking as mne.setup_source_space() instead of a generic mesh decimation. mne.viz.plot_alignment() and mne.viz.Brain should also be faster when plotting many sensors (e.g., whole-head MEG systems), by Eric Larson. (#14016)

• Add a show_zero_line parameter (and 0 keyboard shortcut) to mne.io.Raw.plot() and mne.Epochs.plot() to show a reference line at each channel’s zero, by Clemens Brunner. (#14018)

• Add TimeChange behaviour for mne.viz.plot_evoked(), where vertical time cursor is previewed on hover and marked on click. Also change previous channel names pop-up on click to on hover. Refactor mne.viz.plot_ica_sources() when plotting mne.evoked.Evoked instance, unify under mne.viz.evoked._plot_lines, equipping this plot with hover and click behaviours, by Lifeng Qiu Lin. (#14025)

• When multiple overlays are active (via remove_existing=False in add_data()), the interactive Brain viewer now shows an “Overlay” drop-down in the Color Limits dock panel that lets the user switch which overlay’s color limits and smoothing the sliders control. By Payam Sadeghi-Shabestari. (#14031)

• Make mne.viz.plot_evoked_topo() show the currently select time with a solid vertical line also in the main figure, by Marijn van Vliet. (#14032)

• Added a weighted option to mne.chpi.compute_head_pos() that fits all cHPI coils simultaneously with per-coil weights derived from their goodness of fit and inter-coil distance error, avoiding discontinuous jumps (“teleporting”) in the estimated head position caused by coils switching in and out of the best-three subset (#11330). The default will change from False to True in version 1.14; until then, leaving it unset emits a FutureWarning. By Eric Larson. (#14034)

• More complete type checking has been implemented for mne.io, plus Epochs and Evoked, by Eric Larson. (#14036)

• Make the vertical (channel) and horizontal (time) scrollbar handles in the 'matplotlib' browser backend draggable, by Clemens Brunner. (#14040)

• The colorbar in the mne.viz.Brain viewer now automatically switches to scientific notation for tick labels, and has more tick-label spacing. The Help dialog now uses a Qt dialog styled to match mne-qt-browser’s help dialog instead of a matplotlib popup, and the “Press ? for help” status bar message is now clickable, and “Help” text was removed from the top left of the GUI by Payam Sadeghi-Shabestari. (#14054)

• Added support for newer recording software in mne.io.read_raw_nirx(), by Emma Bailey. (#14055)

• Added type annotations to public members of mne/io, mne.Evoked, and mne.Epochs, by Eric Larson. (#14056)

• Sped up mne.stats.spatio_temporal_cluster_1samp_test() by graph reduction and more efficiently applying sign-flips, by Eric Larson. (#14057)

• mne.viz.plot_ica_components() and mne.preprocessing.ICA.plot_components() now set the window title to “Independent Components” instead of adding a suptitle and append the plotted component range when it is contiguous, by Clemens Brunner. (#14058)

• Added regularize="in_argmax" to mne.preprocessing.maxwell_filter() to allow component selection more similar to that of MaxFilter™-2.2, by Eric Larson. (#14059)

• Speed up and slightly improved the accuracy of mne.make_forward_solution(), by Eric Larson. (#14061)

• The colorbar in the mne.viz.Brain viewer now shows tick marks aligned with its labels, and its title now defaults to the active overlay’s key (see add_data()). The interactive viewer can show vertex coordinates and camera orientation, toggled via the v key or a new toolbar button, by Payam Sadeghi-Shabestari. (#14062)

• When reading EGI MFF files with mne.io.read_raw_egi(), passing event_key='cel#' now returns compound event codes (e.g. stm+_1 and stm+_2) so that trials from different conditions can be distinguished via raw.event_id, by Pragnya Khandelwal and Scott Huberty. (#14086)

• On Linux, mne.sys_info() now includes the detected windowing system (Wayland or X11), by Clemens Brunner. (#14099)

• Added surf and cortex controls to the mne.viz.Brain GUI, by Payam Sadeghi-Shabestari. (#14102)

• Add source_id as an optional parameter in mne.viz.ui_events.VertexSelect, by Lifeng Qiu Lin. (#14126)

• Add mne.preprocessing.remove_fmri_gradient_artifact() and the mne.preprocessing.GradientRemover class to remove the fMRI gradient (imaging) artifact from EEG recorded during functional MRI using average artifact subtraction (AAS, [1]), along with a new tutorial (Removing the fMRI gradient artifact) demonstrating the full workflow, by Peter Molfese and Joshua Teves. (#14127)

• Add the mne.datasets.lite_data.data_path() fetcher for the curated data subset used by the JupyterLite browser documentation, by Natneal B. (#14128)

• Allow mne.compute_covariance() to estimate spatial covariance from an mne.Evoked by treating time samples as observations, by Hamza Abdelhedi. (#14129)

• Add a vtk.js drawing backend for MNE’s 3D renderer, selected with mne.viz.set_3d_backend("jupyterlite_notebook") and used by the JupyterLite documentation where VTK cannot load, by Natneal B. (#14144)

• Added a trace-list sidebar to the mne.viz.Brain GUI, replacing legend by Payam Sadeghi-Shabestari. (#14149)

• 3D plot windows (e.g., mne.SourceEstimate.plot()) now follow light/dark mode switches of the operating system (macOS only for now) while they are open, by Eric Larson. (#14151)

• mne.io.read_raw_egi() now reads bad channels from categories.xml (when present in the MFF directory) and populates info["bads"] with channels marked exclusion="badChannels" in NetStation. By Pragnya Khandelwal. (#14156)

• Added a rank parameter to mne.preprocessing.Xdawn for using Xdawn with rank-deficient data without regularization, by Eric Larson. (#14182)

• Add a block parameter to mne.viz.plot_source_estimates(), mne.SourceEstimate.plot() and mne.VolSourceEstimate.plot_3d() to halt execution until the figure is closed, by Cedric Conday. (#14185)

• Add mne.viz.clear_3d_figure() to remove all actors from a 3D figure so that it can be reused for another plot, by Eric Larson. (#14197)

• Speed up reading by optimizing small read paths, by Bruno Aristimunha (#14212)

• Speed up repeated preloading of any file-backed Raw with preload="auto", which persists decoded data in a copy-on-write cache below mne.set_cache_dir(), by Bruno Aristimunha. (#14216)

• Add support for applying selected SSP projectors and reconstructing projected data in sensor space on Raw, Epochs, and Evoked objects via apply_proj(projs=...) and the new public reconstruct_proj(projs=...) method, by Hamza Abdelhedi. (#14220)

• The auto-picked peak vertex and its activity trace in the mne.viz.Brain GUI now follow the active overlay, by Payam Sadeghi-Shabestari. (#14222)

• Add exclude parameter to mne.io.Raw.get_data(), mne.Epochs.get_data(), and mne.Evoked.get_data() by Carina Forster. (#14225)

• Projection reconstruction can now use an explicit Forward model and standard MNE rank handling, by Hamza Abdelhedi. (#14235)

• Speed up preloading Raw data and reading uniform EDF and BDF data by avoiding unnecessary setup and using pure-NumPy record-stride decoders, by Bruno Aristimunha. (#14237)

• Speed up mne.io.read_raw_brainvision() by reading binary data in cache-sized blocks, by Bruno Aristimunha. (#14241)

• Speed up channel selection and projector construction by replacing linear channel-name lookups with dictionaries, by Bruno Aristimunha. (#14245)

• Speed up reading large Artemis123, Curry, EEGLAB, Eximia, FIL and NSx files by reading them in cache-sized blocks, by Bruno Aristimunha. (#14246)

• Sped up various evoked plotting functions by taking advantage of blitting, and mne.Report image embedding by compressing rendered pixels directly, by Eric Larson (#14249)

• Speed up mne.io.read_raw_egi() on simple-binary files by reading the event channels in blocks instead of one sample at a time, by Bruno Aristimunha. (#14250)

• Speed up mne.io.read_raw_persyst() and mne.io.read_raw_nihon() by decoding data in cache-sized blocks rather than materializing the whole request, by Bruno Aristimunha. (#14251)

• Speed up mne.io.read_raw_fif() on integer-format files by normalizing the byte order of each data buffer before calibrating it, by Bruno Aristimunha. (#14252)

• Speed up windowed reads and epoching of MEF3 files by reusing the session across reads instead of reparsing its metadata each time, by Bruno Aristimunha. (#14254)

• Speed up reading large KIT files by reading them in cache-sized blocks, by Bruno Aristimunha. (#14255)

• Added a only_if_changed=True parameter to mne.Report.save() to skip writing an already up-to-date file, by Eric Larson. (#14261)

Other changes

• Remove numpy.ndarray from colormap parameter in mne.viz.plot_source_estimates() to solve documentation conflict , by new contributor Lifeng Qiu Lin. (#13797)

• Added cross-references to common modification methods in mne.Info docstring by new contributor Athish M. (#13811)

• Add notes to set_eeg_reference() and Setting the EEG reference about the correct procedure to create an average reference, by new contributor Deep Kaur and new contributor Farzah11. (#13846)

• Add Zuna to related software list by new contributor Jonas Mago. (#13849)

• Fixed typos in error messages when reading measurement info, by new contributor Seth Bashford. (#13881)

• Fixed the fun shape description in mne.Epochs.apply_function()’s docstring, by new contributor Bhargav Kowshik. (#13891)

• Batch and vectorise classifier estimation and scoring in mne.decoding.GeneralizingEstimator.score() for scoring=None, "accuracy", "balanced_accuracy" and "roc_auc", by new contributor Mathias Sablé-Meyer. (#13909)

• Add MNE-RT to related software ecosystem and roadmap by new contributor Payam Sadeghi-Shabestari. (#13950)

• Converted the remaining percent-style string formatting to f-strings in internal code, by new contributor Théo Castillo. (#13974)

• Add AI policy pointer by new contributor sametseckın. (#14037)

• Document how to check and tune the OpenBLAS thread count when decomposition-heavy operations are unexpectedly slow, by new contributor Daria Agafonova. (#14064)

• Correct typo in comment by new contributor Samarth D N. (#14117)

• Fix hemisphere generation in test_vertex_to_mni_fs_nibabel to cover both by new contributor Marcus Campbell. (#14183)

• Avoid pytest 10 deprecation warnings when parametrizing scikit-learn estimator checks, by new contributor Deepesh Sonar. (#14283)

• Adopting the diataxis framework by improving the phantom dipole tutorial, by Carina Forster. (#13584)

• Reorganized EDF submodule structure by moving private functionality into the public mne.io namespace to improve module consistency and avoid exposing private modules in the Credits page, by Aniket Singh Yadav. (#13673)

• Document onset, duration, description, and ch_names attributes of mne.Annotations, by Famous Raj Bhat. (#13680)

• Made evoked.plot() instantiate MNELineFigure when it creates its own figure, aligning this path with the ongoing 2D plotting figure-class refactor discussed in #7751, by Pragnya Khandelwal. (#13795)

• Adding checks in mne.stats.permutation_t_test() to warn the user when provided with bad input values, by Himanshu Mahor. (#13813)

• Added regression coverage for colocated triaxial OPM topomap handling in plot_joint() and ICA.plot_components(), by Pragnya Khandelwal. (#13842)

• Added a helper to split colocated OPM overlap sets into radial and tangential channel groups, and updated topomap regression coverage to use shared triaxial OPM fixtures, by Pragnya Khandelwal. (#13847)

• Convert non-empty drop_log entries from numpy.str_ to regular Python strings, by Clemens Brunner. (#13923)

• Refactor the EGI MFF reader to use mffpy for metadata parsing, improving robustness and maintainability. This includes a shim for PNS sensor parsing to bypass an upstream bug in mffpy and resolves channel naming inconsistencies, by Pragnya Khandelwal. (#13945)

• Replace the internal binary-block EEG sample reader in mne.io.read_raw_egi() with mffpy-backed reading via mffpy.Reader.get_physical_samples_from_epoch, removing manual block-header parsing and np.fromfile calls, by Pragnya Khandelwal. (#13973)

• Replace defusedxml-backed implementations of _get_ep_info and _get_signalfname in the EGI MFF reader with mffpy-backed parsing via mne.io.read_raw_egi(), by Pragnya Khandelwal. (#13991)

• Replace defusedxml-backed _get_gains with mffpy-backed parsing via mne.io.read_raw_egi() and remove the internal _extract helper, fully eliminating defusedxml from the EGI MFF reader, by Pragnya Khandelwal. (#14007)

• Add table with all available config options to Configuring MNE-Python, by Clemens Brunner. (#14010)

• Remove _get_eeg_calibration_info from the EGI MFF reader and delegate per-channel GCAL scaling to mffpy’s native calibration, simplifying EEG channel calibration in mne.io.read_raw_egi(), by Pragnya Khandelwal. (#14011)

• Replace the manual PNS binary block reader in mne.io.read_raw_egi() with mffpy-backed reading via get_physical_samples_from_epoch, by Pragnya Khandelwal. (#14030)

• Replace manual binary block reader _get_blocks with mffpy.Reader API in _read_mff_header, removing low-level EGI binary parsing in favour of the existing mffpy dependency, by Pragnya Khandelwal. (#14043)

• Remove dead placeholder keys gain, bits, and value_range from _read_header in egimff.py; these were hardcoded to zero and unused since MFF calibration is handled via CAL_SCALES, by Pragnya Khandelwal. (#14047)

• Move default values out of parameter type descriptions and remove optional from docstrings, by Daria Agafonova. (#14100)

• Decomposition-heavy operations such as mne.preprocessing.maxwell_filter(), mne.compute_covariance() and mne.preprocessing.ICA.fit() now limit how many BLAS threads they use, which is often much faster; see A function uses multiple CPU cores even though I didn’t tell it to. Why? to control this yourself, by Eric Larson. (#14103)

• Add a build script and make target that produce the MNE wheel the JupyterLite browser kernel installs, by Natneal B. (#14135)

• Speed up mne.io.read_raw_egi() for MFF files by preserving the idx slice in _read_segment_file, avoiding a costly np.take copy on large data buffers (~60% reduction in read time for large recordings with preload=True). By Pragnya Khandelwal (#14137). (#14137)

• Add an EGI section to the Getting impedances from raw files gallery example demonstrating how to read per-channel GCAL/ICAL calibration data from an MFF info1.xml file via mffpy and visualize impedances on a topographic map. By Pragnya Khandelwal (#14143). (#14143)

• The Representational Similarity Analysis example is now built and rendered in the documentation instead of being skipped, by Eric Larson. (#14196)

• Add categories to the Related Python packages list, by Laetitia Grabot, Daniel McCloy. (#14232)

Authors

• Alejandro Santos Mayo+

• Aman Srivastava

• Amirali Moradniaei+

• Aniket Singh Yadav

• Anton Karpov+

• Antonio Elias Sánchez Soto+

• Apoorva Verma+

• Athish M+

• Ayushi Satodiya+

• Baris Talar+

• Beliz Sertcan Gökmen+

• Ben Tang+

• Bhargav Kowshik+

• Bruno Aristimunha

• Carina Forster

• Cedric Conday+

• Chetan Gohil

• Christian Brodbeck

• Clemens Brunner

• Daniel McCloy

• Daria Agafonova+

• Deep Kaur+

• Deepesh Sonar+

• Emma Bailey

• Eric Larson

• Erica Peterson

• Famous Raj Bhat+

• Farzah11+

• Frankie Robertson+

• Hamza Abdelhedi

• Himanshu Mahor

• Jacob Woessner

• John Griffiths+

• Jonas Mago+

• Julien Dubois+

• Kalle Mäkelä

• Laetitia Grabot

• Lifeng Qiu Lin+

• Luca Kaemmer+

• Marcus Campbell+

• Marijn van Vliet

• Martin Oberg

• Mathias Sablé-Meyer+

• Mathieu Scheltienne

• Michael Straube

• Mingjian He

• Natneal B

• Nikolai Kapralov

• Pavel Popov+

• Payam Sadeghi-Shabestari+

• Peter Molfese

• Piotr Durka+

• Pragnya Khandelwal

• Qian Chu

• Raphaël Bordas+

• Robert Luke

• Samarth D N+

• Scott Huberty

• Seth Bashford+

• Seyed Yahya Shirazi

• Sharif Haason+

• Sina Esmaeili+

• Stefan Appelhoff

• TIM ANDERSON

• Thomas Binns

• Théo Castillo+

• Tom Ma

• Varun Kasyap Pentamaraju+

• Vincent Gao+

• Virginie van Wassenhove+

• Zahra M. Aghajan+

• sametseckın+