PNPL Competition 2026

Current

Word Classification from MEG Brain Recordings

OverviewTaskTimelineRulesParticipateLeaderboardPublicationsOrganisersFAQ
Now open · Submissions 15 Jul 2026

The PNPL Competition is back. Our resources and first tutorial are live now so you can start experimenting today — submissions open 15 July and close 15 October 2026 (AoE).

The 2026 PNPL Competition builds on the success of our inaugural year with a more ambitious task and a significantly expanded dataset. This year, participants will tackle Word Classification — predicting which word a subject is hearing from MEG brain recordings.

What's New

New Task

Word Classification — decode which of 50 words a subject is hearing from MEG data

New Dataset

Introducing the new LibriBrain100 dataset, now with over 100 hours of MEG data

Two Tracks

Go Deep on a single subject or Broad across 32 — enter either or both

Three Months

Submissions run 15 July – 15 October 2026 (Anywhere on Earth).

The Dataset: LibriBrain100

LibriBrain100 is a major expansion of the original LibriBrain dataset, bringing the total to over 100 hours of MEG data. It pairs the deeply-sampled subj0 recordings behind the Deep track — spanning audiobooks, phonetically balanced speech corpora (TIMIT, MOCHA-TIMIT), and narrative podcasts — with varying amounts (~10–40 minutes each) from 32 additional subjects behind the Broad track.

This multi-subject dataset opens the door to cross-subject generalisation — a critical challenge for practical brain-computer interfaces. How much data does it take to reach useful performance on a new subject?

Overview over the LibriBrain100 dataset.
Scatter plot comparing LibriBrain100 with other MEG datasets on number of subjects and maximum hours per subject. LibriBrain100 sits in the upper-right quadrant — many subjects and many hours per subject — while datasets like MEG-MASC, MOUS, and Le Petit Prince have more subjects but far fewer hours each.
LibriBrain100 compared with other public MEG datasets. The original LibriBrain went deep on a single subject; LibriBrain100 keeps that depth while adding breadth across 33 subjects.
Task Details →View 2025 Results