Meta AI introduces NeuralBench — a framework for testing brain activity models

Meta AI released NeuralBench — a unified framework for benchmarking models that analyze brain activity recordings. Simultaneously, NeuralBench-EEG v1.0 was released — the largest open electroencephalography dataset in history, covering 36 diverse brain signal processing tasks and 94 individual datasets, compiled from data of 9,478 subjects with a total of 13,603 hours of high-quality EEG recordings.

What is NeuralBench

NeuralBench provides a single standardized interface for fair testing of 14 different deep learning architectures on identical data. This solves a fundamental problem that has plagued NeuroAI researchers for decades: previously, each laboratory used its own datasets, applied its own EEG signal processing methods, and chose its own evaluation metrics. Because of this, results could not be objectively compared between groups. The framework covers various types of EEG tasks — from signal classification and artifact detection to predicting cognitive states and emotions. Each of the 36 tasks has clearly defined evaluation metrics that eliminate subjectivity in interpreting results.

Why unification was needed

Before NeuralBench, NeuroAI remained deeply fragmented. Different research groups applied different approaches, used different data processing tools, and different ways of evaluating models. This seriously hampered reproducibility of results, objective comparison of methods, and tracking overall progress in the field. A unified approach enables:

• Quickly evaluate new architectures without a month of preliminary engineering work
• Fairly compare models from different labs under identical conditions
• See overall field progress on a single scale
• Develop NeuroAI as an engineering discipline with common standards
• Transfer knowledge between applications — from diagnosis to brain-computer interfaces

For whom this is critical

The framework is important for neuroscientists who want to apply AI to EEG data and for ML engineers interested in neuroscience. Companies developing brain-computer interfaces — from thought-controlled prosthetics to post-stroke recovery systems — can now validate models on a recognized benchmark. This will increase investor and medical regulator confidence in new technologies. The openness of the dataset is of enormous importance. It is available to everyone for free, so any team can begin work on NeuroAI applications without purchasing expensive EEG data collection equipment.

Historical precedent

NeuralBench is to NeuroAI what ImageNet was to computer vision. When a public benchmark with a large dataset appeared in 2010, it launched a golden age of computer vision development. Architectures improved, objective standards emerged, and it became clear which approaches and methods actually work. The same will happen with brain analysis. A shared benchmark accelerates the industry: researchers get a clear goal, companies invest confidently, quality improves for all participants.

What this means

NeuralBench could significantly accelerate the transition of neural interfaces from laboratory prototypes to clinical practice and commercial applications. Doctors will get objective performance metrics. Investors will see a standardizing market. Researchers will be able to focus on innovation rather than data compatibility and formats. This is a rare moment: NeuroAI is mature enough for a useful benchmark, but still young enough that unification could significantly accelerate development for years to come.