Skip to content
NewA calmer, faster ZeroHour
StoryModel releaseinfo

Block unveils Dessa, a speech-only voice model for safety testers

3 sources · 3 articlesFirst reported Updated ()
What's new

First merged summary.

Merged summary · x-ai/grok-4.7 · rewritten as coverage arrivesSummaries by Vani AI

Block’s Dessa speech model is not public; the company is first offering early access to safety and security researchers.

Jack Dorsey’s Block announced Dessa, a full-duplex speech-to-speech model the company says was trained from scratch only on human conversation, not written text. Block says that design lets it use emotion, sarcasm, laughter, and pauses in real time, and it published sample chats and voice clips, including awkward exchanges and a comparison with OpenAI’s GPT-Live-1. The company says Dessa runs locally on a MacBook Pro M4 Max or better; The Decoder adds that it was trained on one NVIDIA GB200 NVL72, while TechCrunch says it runs locally on one NVIDIA GPU, including that MacBook. Dessa is not publicly available. Early access is limited to a small group of AI safety and security researchers, including tests of cases where a voice agent could deceive a listener.

Key facts
  • Block claims Dessa is the first model trained entirely on human speech, with no text.
  • The model can listen and talk at the same time.
  • One published chat shows the model initially calling the speaker Ronaldo.
  • Researchers can request access through a signup form.
  • Block cites Hume EVI, Moshi, and ChatGPT Advanced Voice Mode as earlier voice efforts.
Explain this AI news with Vani AIWhat was announced, what is new and why it matters, in a new chat

Timeline

dated events in the reporting · oldest first
  1. Oct 23, 2026Researchers posted Block’s Dessa announcement. · TechCrunch

What each outlet adds

beyond the shared facts
  • The VergeSample chats, the Ronaldo mix-up, and a GPT-Live-1 comparison.
  • The DecoderTraining on one NVIDIA GB200 NVL72 and a researcher signup form.
  • TechCrunchInvitation to test voice-agent deception and the October 23 post date.
Vendors
Block
Products
Square
Organizations
BlockOpenAI
AI models
DessaGPT-Live-1

Coverage

oldest first · one row per article · UTC
  1. · 2d ago The Verge · AIFirst report

    Block unveils Dessa, a speech-only conversational AI model

    Block says it runs locally on a MacBook Pro M4 Max or better and will give safety researchers early access.

  2. · 1d ago The Decoder

    Block unveils Dessa, a speech-only model not yet public

    It runs locally on a MacBook Pro M4 Max, was trained on one NVIDIA GB200 NVL72, and Block wants security testers.

  3. · 1d ago TechCrunch · AI

    Block releases Dessa, a local full-duplex speech model

    Dessa is not publicly available, and Block has asked safety researchers to test it first.