Oct 27, 2026

Dessa: Block's voice AI learned to talk without reading a word

Block's new research model Dessa learned to speak only from recordings of human conversation, with no text and no pretrained models. It's a research preview, with no public release yet.

Sir Toastimir

Footage: Block

What Block announced

Block, Jack Dorsey's company, has shown Dessa, a voice model from a small lab inside Block. The lab was founded by former members of the Toronto AI startup Dessa and is led by Ragavan Thurairatnam.

How it's different

Block says it trained Dessa from scratch on nothing but recordings of human conversation: no text and no pretrained models. Most voice assistants are built on top of text language models. Block says Dessa learned language from hundreds of times fewer words than large language models use.

Dessa is full-duplex: it listens while it talks. That lets it react the way people do, for example laughing mid-sentence when something is funny. Block says it natively hears and speaks emotion and sarcasm.

Where it runs

According to Block, Dessa runs locally on a MacBook Pro with an M4 Max chip or better, faster than real time.

What we don't know yet

This is a research preview. Block is giving a small group of AI safety researchers early access, and there is only a waitlist for everyone else: no public app, API or model weights. The claims above come from Block; we haven't been able to try the model ourselves.

Visuals in the video

All footage is from Block's official Dessa announcement videos.