Block's New Dessa Voice Model Runs on a MacBook M4 Max

A single Nvidia GPU rack is all Block says it needed to train a voice model the company claims can run on a consumer-grade MacBook M4 Max, according to TechCrunch. The model, called Dessa, comes out of Block's research lab under founder Jack Dorsey, and the company is positioning it as a direct challenge to text-driven voice assistants that dominate the market today.

The move matters because most commercial voice AI still routes speech through text before generating a reply, a pipeline Block argues strips out tone, hesitation, and emotion before the system ever responds. Dessa skips that step. It is a full-duplex model, meaning it can listen and speak at the same time, a design Block compares to OpenAI's GPT-Live model released earlier in 2026, per TechCrunch's reporting.

What is Block's Dessa voice model?

Dessa is trained entirely on human speech rather than paired with a separate large language model for knowledge retrieval, TechCrunch reported. Instead, the model draws its responses from the conversational data embedded in the speech itself. Block says that architecture is what lets Dessa produce laughs, pauses, and cries without relying on manually inserted emotion cues.

"Interacting with current voice AI can just plain suck. It turns your voice into text and throws away an incredible amount of information text alone can't carry. The most powerful AIs in the world can't hear how you feel. We think that's ridiculous, leads to a worse experience, and is potentially dangerous," Block said in a blog post cited by TechCrunch.

Why does Dessa sound more human than other AI voices?

Block contrasts its approach with tools such as ElevenLabs v4 and Fish Audio's Drama 3, which TechCrunch says typically use voice tags to steer emotional delivery after the fact. Dorsey wrote on social media that talking with Dessa "feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence," a claim TechCrunch notes is not fully verifiable since it is unclear whether Dessa's emotional output is still shaped by some form of internal tagging.

Researchers tied to the project posted example conversations to illustrate the point. In one, the bot said it was tired from work and sounded tired while saying it, according to TechCrunch. In another, a Block team member named Cara played a guessing game with the model, which correctly guessed an answer on its first try. One researcher, Ragavan Thurairatnam, described the release on social media as "a GPT-3 moment for speech."

Can Dessa run on a MacBook M4 Max?

Yes. Block told TechCrunch the model can run locally on consumer-grade hardware, specifically naming the MacBook M4 Max, rather than requiring cloud infrastructure. That local-run capability is central to Block's pitch: a speech model trained on one Nvidia GPU rack that can then operate on a laptop chip, cutting the dependency on remote servers that most voice assistants still need.

How does Dessa compare with ElevenLabs and OpenAI's GPT-Live?

TechCrunch frames Dessa as entering a field that already includes OpenAI's full-duplex GPT-Live model and commercial voice platforms such as ElevenLabs, which markets tools built for production voice applications and can be tested directly through ElevenLabs. The distinction Block is drawing is architectural: where ElevenLabs and Fish Audio lean on text-plus-tag pipelines, Dessa is pitched as speech-native from the ground up. Whether that distinction holds up under independent testing has not yet been verified outside Block's own demonstrations, per TechCrunch's reporting.

What's next for Block's voice AI?

Block has not detailed a commercial release timeline for Dessa in the material TechCrunch published, and the examples so far come from the company's own posts rather than third-party benchmarks. For consumers curious about where consumer-facing voice-persona apps are headed, Smorgi Apps' AlterEgo explores a similar space of AI conversational personas, while a planned app called Artiglio is coming soon from the same developer. Readers tracking the broader voice-AI market can follow the original TechCrunch account for updates on independent testing of Dessa's claims.