Leaderboard
Bits per second (BPS) measures how much information a person with a brain-computer interface (BCI) sends by moving and clicking a cursor on a grid of targets. The record stayed near 4 BPS for seven years, and Neuralink participants have pushed it to 11.32 since 2024. Each row below names the person, the device, the task, and the source.
Current record, set by Neuralink participant P15 on Webgrid and reported on October 1, 2026.
Best pre-Neuralink grid result, by BrainGate participant T5 in a study published in February 2017.
Score of the median Neuralink participant on Webgrid, according to Neuralink in October 2026.
How the record was set: Neuralink's 11.32 BPS record and the 50,000 hours of data behind it.
The chart plots only grid-selection results, because those tasks measure the same thing. Results from other tasks are in the leaderboard and marked as not directly comparable.
Each point is a new best score. Hover over a point for details.
| Date | Who | BPS |
|---|---|---|
| Nov 2015 | T6, BrainGate | 0.93 |
| Feb 2017 | T5, BrainGate and Stanford | 4.36 |
| Early 2024 | Noland Arbaugh, Neuralink | 4.6 |
| May 2024 | Noland Arbaugh, Neuralink | 8.0 |
| Jul 2024 | Noland Arbaugh, Neuralink | 9.51 |
| By Nov 2025 | Participant 8, Neuralink | 10.39 |
| Oct 2026 | P15, Neuralink | 11.32 |
Each person's best published result, ranked by BPS. The last column says whether the task can be compared with Neuralink's Webgrid test.
| Rank | Who | Device | BPS | Task | Date | Source | Compared with Webgrid |
|---|---|---|---|---|---|---|---|
| 1 | P15Neuralink | N1 implant | 11.32 | Webgrid | Oct 1, 2026 | Neuralink | Same task |
| 2 | Participant 8Neuralink | N1 implant | 10.39 | Webgrid, 40 by 40 grid | By Nov 2025 | Neuralink Webgrid page | Same task |
| 3 | Noland ArbaughNeuralink, patient 1 | N1 implant | 9.51 | Webgrid, 45 by 45 grid | Jul 2024 | Neuralink Webgrid page | Same task |
| 4 | Three patientsStairMed, Shanghai | Ultra-flexible electrode implant | 6.3 | Cursor control, task not published (reported as 380 bits per minute) | Jul 30, 2026 | Global Times | Not comparable |
| 5 | T5BrainGate and Stanford | Two 96-channel Utah arrays | 4.36 | Grid selection, best block (4.16 average on a 9 by 9 grid) | Feb 2017 | Pandarinath et al., eLife | Similar task |
| 6 | Epilepsy patient, age 19NeuroXess, Shanghai | 256-channel flexible implant | 4.07 | Cursor control with an adapted interface | Apr 2025 | China News Service | Not comparable |
| 7 | Participant with ALSUC Davis and BrainGate | Utah arrays in the speech motor cortex | 2.90 | Cursor target selection and click | Nov 2024 | bioRxiv preprint | Not comparable |
| 8 | T6BrainGate | Utah array | 0.93 | Grid selection | Nov 2015 | Pandarinath et al., eLife, table 1 | Similar task |
Noland Arbaugh also scored 4.6 BPS in his first research session in early 2024 and 8.0 BPS by May 2024, both reported by Neuralink. Neuralink says three participants beat 10.39 BPS with its new decoders in 2026, and it has published only P15's score. Neuralink has not named P15, and its Webgrid page credits 10.39 to "our eighth clinical trial participant".
Webgrid multiplies the bits in each choice by the net rate of correct selections, which is correct clicks minus wrong ones. A bigger grid packs more bits into each click, and Neuralink's records used 40 by 40 and 45 by 45 grids, far finer than the 6 by 6 and 9 by 9 grids of the 2017 BrainGate study.
Some studies report a session average and others report the best single run. T5 averaged 4.16 BPS on the 9 by 9 grid and peaked at 4.36, while Neuralink lists each participant's best Webgrid score.
The StairMed and NeuroXess figures come from news reports that do not describe the grid or the scoring, so they appear in the leaderboard without a place on the chart.
Speech and handwriting BCIs report words or characters per minute, which do not convert to grid BPS.
Neuralink's figures come from its own update posts and its Webgrid page, read through Internet Archive snapshots where the page has since changed. Academic figures come from the published papers. Company figures have not appeared in a peer-reviewed journal. Anyone can try the same test with a mouse on Neuralink's Webgrid.
Neura Pod is independent and has no affiliation with Neuralink. Neuralink devices are investigational and not approved by the FDA.
