📅 Oct 25, 2026
Block unveiled Dessa, a voice model it says was trained only on human conversation — no text, no pretrained models.
LexxWire
Most voice assistants turn speech into text first. Dessa skips that step entirely: audio in, audio out.
LexxWire
Dessa is full-duplex, so it can cut in, laugh mid-sentence, and backchannel with a 'yeah' or 'got it' like a real person.
LexxWire
Block says Dessa runs locally on a MacBook Pro M4 Max or better, with inference faster than real time.
LexxWire
Dessa is a research preview behind a waitlist, with early access for a small group of AI safety researchers first.
LexxWire
Thanks for reading!