Neuralink Says an ALS Patient Hit 75 Words a Minute With Its Speech Implant in Week One. It's the First Real Number From the VOICE Trial, and It Comes From One Person.

Honest fence (read first): This is a company-reported result from one participant in a small early trial, shared in an X post and a conference abstract. There's no peer-reviewed paper yet. The device is investigational and isn't approved by the FDA or any other regulator.

Editorial Thesis

On Wednesday, Oct 28, 2026, at 4:02 PM ET, Neuralink posted the first hard numbers from its speech trial: "One participant achieved 75 WPM at 5.8% WER on a 125k-word vocabulary in their first week." In plain English, one person with ALS turned attempted speech into text at 75 words a minute, and the system got about 94 of every 100 words right, drawing on a vocabulary of 125,000 words. A follow-up post said Dr. Nader Pouratian, the trial's principal investigator, presented the data that day at the annual meeting of the Northeast ALS Consortium (NEALS).

Until now, every VOICE update was a video clip: a participant speaking to his wife in a rebuilt voice, a demo, a story. This is the first time Neuralink has put a speed and an error rate on the record. That's real progress in what the company is willing to show. It's still one person's first week.

Honest Verdict

Believe the number exists and that Neuralink stands behind it. It's in a company post and a conference abstract with a named investigator. Believe it's fast for this field: on paper, 75 words a minute at 5.8% errors on a 125,000-word vocabulary is quicker than the best published results on that vocabulary size.

Don't read it as "ALS patients can talk at normal speed again." Normal conversation runs at about 150 to 160 words a minute. Neuralink didn't say whether 75 was a peak or an average, whether the sentences were prompted or freely chosen, or how the result held up after week one.

What Neuralink actually said

• The claim (X, Oct 28, 4:02 PM ET): "Early Neuralink speech restoration data: We are enabling speech to text and voice at conversational rates in participants with ALS. One participant achieved 75 WPM at 5.8% WER on a 125k-word vocabulary in their first week."
• The venue: "Presented today at the 2026 Annual NEALS Meeting by Dr. Nader Pouratian," with a link to the abstract in the meeting proceedings, published in the journal Muscle & Nerve. Neuralink's own Updates page lists it as "Real-Time Communication Restoration in ALS with a Fully Implanted Speech Brain-Computer Interface."
• The fine print: "Neuralink devices are investigational and have not been approved by the FDA or other regulatory authorities."

"WPM" is words per minute. "WER" is word error rate, the share of words the system gets wrong. A 5.8% WER means about 94 of every 100 words came out right.

What the abstract adds

We couldn't open the abstract's full text from our system, so treat this part as second-hand. A public summary of the abstract by the ITR Atlas, a site that tracks brain-interface speeds, says it covers three adults with ALS:

• Participant 3: 75 words a minute at 5.8% WER on the 125,000-word test, reached in under a week after implant, with what the summary calls the first-in-human 3,072-channel version of Neuralink's N1 implant.
• Participant 2: 54 words a minute at 6.6% WER on the same test, with a 1,020-channel implant.
• Participant 1: 45 words a minute at 7% WER, but on a much smaller 220-word vocabulary, so that one isn't comparable.

If that summary is right, the headline number comes from the newest, highest-channel implant, and the next-best participant was well behind it. That's worth knowing before anyone calls 75 the "typical" result.

The trial behind it

The data comes from VOICE (ClinicalTrials.gov NCT07224256), an early feasibility study. Per the registry:

• It started on Oct 3, 2025, at UT Southwestern Medical Center in Dallas, the only listed site.
• It plans to enroll about 6 people, aged 22 to 75, with severe speech loss and limited arm and hand use.
• Its primary outcomes are safety: device- and procedure-related adverse events at 12 months.
• Speech performance is a secondary measure, recorded as the monthly maximum of correct words per minute.

That last point matters. The trial's own yardstick is a monthly best, not an average. Neuralink received FDA Breakthrough Device designation for speech in May 2025. That speeds up reviews. It isn't an approval.

How 75 compares

These comparisons are rough, because each lab uses different people, implants and tests. Inside BCI pulled the published benchmarks together:

On paper, Neuralink's figure is faster than the large-vocabulary results and far more accurate than Stanford's 2023 number. UC Davis's sustained accuracy over months is still the better proof of daily reliability. Neuralink's own speech-trial page says its "median participant decodes speech at roughly 50 words per minute," which lines up better with the 54 reported for participant 2 than with the 75 headline.

Why it matters

For someone with ALS, speed is the difference between a tool and a conversation. Eye-gaze typing, the common fallback, runs at about 8 words a minute by Neuralink's own estimate. Getting to 50 to 75 words a minute, with few errors, starts to feel like talking instead of spelling.

It also gives the field a number to check. Paradromics and other implant makers are chasing speech too, and Inside BCI notes Paradromics hasn't published speed or error figures for its first speech participant. Neuralink has now set a public bar that it will be measured against, including by its own next update.

What Neuralink hasn't said

• How many VOICE participants are decoding speech today.
• Whether 75 words a minute was a single best session or a sustained average.
• Whether the test sentences were prompted or the participant's own words.
• How the error rate was scored.
• How results held up after the first week.

The video

Farzad's Saturday video ("Elon's Neuralink Just Made An Insane Breakthrough," 16:13) walks through the VOICE trial and the 75 words-a-minute post starting at 7:12, then covers Neuralink's 72-hour speech hackathon and Musk's Blindsight post. It's a creator explainer, not a primary source, and Farzad says he's a Tesla shareholder. For a broader, non-news primer on how brain implants work, Cleo Abram's Oct 30 video is a good alternate.

What would change our read

• A peer-reviewed paper with methods, averages over months and all participants.
• A second participant reaching similar speed and accuracy.
• Evidence that the decoder keeps working with little recalibration, as Neuralink's Oct 1 cursor work suggested for typing and pointing.

Bottom line

Neuralink has finally put a speed and an error rate on its speech work, and the numbers are strong: 75 words a minute at a 5.8% error rate on a 125,000-word vocabulary in one person's first week. That's the claim. The proof that matters for patients, sustained averages across several people over months, isn't public yet.