Block announces Dessa, a speech-to-speech model trained from scratch on human conversation

Block, the company led by Jack Dorsey, has announced Dessa, a speech-to-speech AI model that it says was trained "from scratch on nothing but human conversation." The team says Dessa can run locally on a MacBook Pro M4 Max "or better." Early access will go to "a small group of AI safety researchers." The announcement was reported by The Verge on Oct 23, 2026.

What Block says Dessa is

According to the Dessa website, the model is a "true speech-to-speech model, and everything it knows came from human speech." Dessa lead Ragavan Thurairatnam shared one of the demo conversations.

Pricing not yet disclosed. Block has not stated a context window, parameter count or benchmark scores in the material The Verge reported.

Demos

Block published several conversations between Dessa and humans:

• A guessing game played over voice.
• A back-and-forth about Christopher Nolan's films, in which Dessa goes on an unexpected tangent about Bryan Singer's Superman films.
• A person describing a surprise trip to Paris they took Dessa on. The website says Dessa "then answers with what sounds like teary gratitude."
• A conversation between OpenAI's GPT-Live-1 voice model and Dessa, in which Dessa comes across as more natural.

The Verge describes the exchanges as mostly natural but occasionally awkward and uncanny. In the conversation shared by Thurairatnam, the model initially mistakes the speaker for somebody named "Ronaldo."

Why Block built it

The team gives three reasons: "curiosity (aka fun), better understanding between humans and AI, and safety." In its own words:

• "Babies learn to talk without ever seeing a word written down. We wanted to see if a model could too. It can. It's just very hard."
• Speech is "the most natural interface to any intelligence, super or otherwise, and today's voice AI wastes it. It flattens what we say into text and can't keep up with human timing."
• On safety, the team says AI is gaining power and independence while remaining "nearly deaf to the emotion in our voices." Its example is a person in serious pain while a voice AI has no idea and keeps going. The team says Dessa is "a step toward fixing it."

Access

The Dessa team says it will give a small group of AI safety researchers early access to the model. No broader availability, API access or release date is stated.

What this means

Dessa's central claim is architectural: a model that takes in and produces speech directly, with its knowledge drawn only from human conversation rather than written text. The team contrasts this with voice systems that convert speech to text first, which it says loses emotion and timing. The local-run claim, on a MacBook Pro M4 Max or better, sets a hardware bar for on-device use.

The evidence so far is Block's own demos and its own descriptions of them, including the comparison in which Dessa sounds more natural than GPT-Live-1. Access is limited to a small group of safety researchers, so those claims have not been tested outside Block's selected examples. The reported glitches, such as the "Ronaldo" mix-up, show that the shared conversations were not edited to look flawless.