Neuralink hits 11.32-bit/s record with 50,000 hours of brain data

Summary

Neuralink trained its self-supervised neural decoders on over 50,000 hours of intracortical data collected from clinical trial participants. The company claims to have set a new record of 11.32 bits per second in cursor control, noted that the decoders remained stable for weeks, and stated that the calibration frequency has dropped from daily to weekly.

Why it matters

Signal drift was the biggest practical hurdle for brain-computer interfaces intended for daily use. If decoders truly remain stable for weeks and the need for recalibration decreases, the technology shifts from a laboratory curiosity to a usable input device for everyday life. The 50,000-hour dataset places Neuralink in a position of de facto data advantage in the intracortical BCI field, an area with very few competitors.

Highlights

• Reports indicate that calibration time for some participants dropped from approximately 10 minutes per day to approximately 10 minutes per week
• Self-supervised decoders trained on over 50,000 hours of motor cortex spike data
• The results are derived from a company technical update rather than a peer-reviewed publication

Critical take

The 11.32 bits per second figure is explicitly stated as unverified in comparisons with healthy individuals. The entire result is presented as a corporate 'technical update' lacking independent replication, participant numbers, and ablation details. This situation carries the risk that the 'foundation model' framework is outpacing the evidence. Until a peer-reviewed paper including raw data and speed ratios is published, this development appears more like a well-timed PR milestone than a scientific breakthrough.