←Story detail

Trending story

Block's research team launches pure voice model Dessa

1 reports1 story updatesUpdated Oct 28 09:58

Understand this story

AI overview

On October 28, 2026, Block's research team launched the voice model Dessa, whose training process completely excludes text; it can listen and speak simultaneously, interject naturally, and run locally on a MacBook Pro equipped with an M4 Max chip. Phoenix Tech reported the same day, contrasting Dessa with xAI's Grok Voice 3.0 approach: the latter focuses on real-time end-to-end voice interaction, targets sub-second response times as its optimization goal, and handles approximately 20,000 customer service calls per day for Starlink, with about 70% of issues requiring no transfer to a human agent. The report also mentioned that OpenAI updated its voice model GPT-Live in mid-October, adding nine new voices to the API and supporting custom voices described in text. There is currently no independent factual information on further developments regarding Dessa.

AI-generated from public reports

Latest development · Report

Twitter founder Jack Dorsey's voice model Dessa is not trained on text, forming a contrast with the Grok Voice 3.0 approach

Developments

1 confirmed developments

  1. Block research team launches pure voice model Dessa

    Source:Phoenix Technology

Reports and discussion timeline

Browse reports and discussion by time. Their evidence levels remain distinct.

Oct 28

  1. Phoenix Technology

    Twitter founder Jack Dorsey's voice model Dessa is not trained on text, forming a contrast with the Grok Voice 3.0 approach

    Block's research team Dessa has launched a voice model whose training process completely excludes text; it can listen and speak at the same time, interject naturally, and run locally on a MacBook Pro equipped with an M4 Max chip. The article contrasts it with xAI's Grok Voice 3.0, which focuses on real-time end-to-end voice interaction, targets sub-second response times as its optimization goal, and handles about 20,000 customer service calls per day for Starlink, with roughly 70% of issues requiring no transfer to a human agent. The article also mentions that OpenAI updated its voice model GPT-Live in mid-October, adding nine new voices to the API and supporting custom voices described in text, while also discussing the boundaries of voice in terms of information bandwidth and social scenarios.

Related stories

  1. Related storiesBlock releases full-duplex speech model Dessa
  2. Related storiesMistral releases Large 4 weights on Hugging Face
  3. Related storiesApple announced that the iOS 27.3 developer beta will be released next week
  4. Related storiesAnthropic Publicly Submits S-1 Registration Statement for Proposed IPO on Nasdaq