Source details
- Original source
- The Rundown AI
- Published
- 2026-06-29
- Primary topic
- Developer Tools
Why it matters
Coding copilots, APIs, SDKs, open-source releases, dev workflows, and engineering infrastructure. This item originated as a short-form social post, so the context blocks below help expand it into tools, models, and evaluation guides.
What happened
Meta got a brain-to-text decoder to 61% word accuracy, reading raw signals from outside the skull without any implants or surgery. The previous best for reading the brain without surgery = ~8%. It learned from 9 volunteers, who each sat 10 hours inside a brain scanner and typed while the system read along. One AI model read the raw brain signals as they typed, and a language model filled in the meaning. The top volunteer hit 78%, with over half of their sentences came back with one word wrong at most. This is v2 of Brain2Qwerty. v1 was published today in Nature, and only decoded one character at a time. v2 jumped to whole words and their meaning, with Meta releasing the training code for both. And accuracy keeps climbing with more data, meaning the gap with surgical implants might close on scale alone. AI at Meta (@AIatMeta) We’re sharing the next major milestone in our non-invasive brain-to-text decoder research: Brain2Qwerty v2. Building on v1, which was published today in @Nature , Brain2Qwerty v2 is the highest-performing end-to-end pipeline capable of real-time sentence decoding from raw brain signals. It advances beyond character-level performance to decoding words and semantics, enabling accuracy for overall communication. We believe this research has the potential to make a real difference for the millions of people who suffer from brain lesions or disorders that prevent them from communicating. 🧵👇 — https://nitter.net/AIatMeta/status/2071566924803395741#m
What to do next
Jump to coding tools and tutorials to see whether the release changes real implementation speed, reliability, or cost.
Meta got a brain-to-text decoder to 61% word accuracy, reading raw signals from outside the skull without any implants or surgery. The previous best for reading the brain without surgery = ~8%. It learned from 9 volunteers, who each sat 10 hours inside a brain scanner and typed while the system read along. One AI model read the raw brain signals as they typed, and a language model filled in the meaning. The top volunteer hit 78%, with over half of their sentences came back with one word wrong at most. This is v2 of Brain2Qwerty. v1 was published today in Nature, and only decoded one character at a time. v2 jumped to whole words and their meaning, with Meta releasing the training code for both. And accuracy keeps climbing with more data, meaning the gap with surgical implants might close on scale alone. AI at Meta (@AIatMeta) We’re sharing the next major milestone in our non-invasive brain-to-text decoder research: Brain2Qwerty v2. Building on v1, which was published today in @Nature , Brain2Qwerty v2 is the highest-performing end-to-end pipeline capable of real-time sentence decoding from raw brain signals. It advances beyond character-level performance to decoding words and semantics, enabling accuracy for overall communication. We believe this research has the potential to make a real difference for the millions of people who suffer from brain lesions or disorders that prevent them from communicating. 🧵👇 — https://nitter.net/AIatMeta/status/2071566924803395741#m
This AimostAll brief summarizes the linked source so readers can scan AI developments quickly and jump to the original reporting when needed.
