Jack Dorsey, co-founder and chairman of Block, photographed in 2014Photo: Rory Cellan-Jones (cellanr), CC BY-SA 2.0, via Wikimedia Commons

Block, the payments company co-founded by Jack Dorsey, has unveiled Dessa, an experimental voice AI model that the company says was trained entirely on recordings of human conversation, without any text.

Dorsey announced the model on X on 23 October 2026. “We believe voice is the most natural way to interact with intelligence, so we created Dessa, the most authentic full-duplex speech model ever built,” he wrote, adding that talking with it “feels like talking with a person.” The post had drawn more than 500,000 views within hours.

Who built it

According to Block’s announcement page, Dessa comes from a “(very) small AI research lab at Block” founded by former members of Dessa, the Toronto deep learning startup that Block, then called Square, acquired in 2020. The lab is led by Ragavan Thurairatnam and reports directly to Dorsey, Block’s co-founder and chairman.

What “full-duplex” means

Most voice assistants work like a walkie-talkie: you speak, they wait until you finish, convert your words to text, and then answer. A full-duplex model listens and speaks at the same time, the way people do on a phone call. That means you can interrupt it mid-sentence, and it can react while you are still talking.

Block says Dessa is a “true speech-to-speech model”: audio goes in and audio comes out, with no step through text. The company says this lets the model pick up emotion, sarcasm, laughter and sighs that are lost when speech is turned into text, and that it learned from real conversation when to cut in and when to offer short cues such as “yeah” or “got it.”

Claims and availability

Block says Dessa was trained from scratch, with no pretrained models, on a single NVIDIA GB200 NVL72 system, that it responds faster than real time, and that it can run locally on a MacBook Pro M4 Max or better. It also says the model learned language from “hundreds of times fewer words” than large language models need. The company has not published formal benchmarks; its page instead shares live recordings of people talking with the model, including one in which Dessa talks with ChatGPT Live (GPT-Live-1), a third-party model whose developer Block says has not endorsed Dessa. The Verge noted that some of these demo chats can be “a little awkward and uncanny.”

Dessa is not publicly available. Block says it is first giving early access to a small group of AI safety researchers to look for risks it has not anticipated, citing concerns that today’s voice AI cannot hear when a user is in distress.

Sources

79 Views

By HahuPrime News Desk

The HahuPrime News Desk reports on Ethiopia, Eritrea, the Horn of Africa, East Africa, the Red Sea and the Middle East. Our reports separate confirmed facts from claims and analysis, attribute every claim, and list their sources. Corrections: info@hahuprime.com.