AIToday
Large Language ModelsAudio & SpeechAI Business & IndustryGIGAZINE AIPublished: Oct 26, 2026, 13:00 JST

Block unveils Dessa, a speech-only AI model

Send on LINE
Block unveils Dessa, a speech-only AI model

3 Key Points

  1. What happened

    Block announced Dessa, which its team calls the world's first AI trained only on human speech audio, skipping text entirely. It was trained on one NVIDIA GB200 NVL72 and can run locally on an M4 Max MacBook Pro.

  2. Why it matters

    Because Dessa learns from audio alone, it can overlap replies, backchannel and laugh mid-conversation, which the team says makes talking with it feel like talking with a person.

WHO IT HITSBusinesses building voice assistants or call-center automation may eventually get a model that handles natural interruptions and emotional nuance, though they cannot access Dessa yet since no API or model is public.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

Dessa comes from a team with a long history: its members are from Dessa, a Toronto startup that has researched AI since 2012, and they now work as a small AI research group inside Block, reporting results directly to CEO Jack Dorsey. Dorsey framed the launch around voice as the most natural way to interact with intelligence, describing conversations with Dessa as full of emotion, nuance, pauses and a natural cadence.

The design choice sets Dessa apart from voice systems built as extensions of text-generating models, which strip out intonation and emotion that text cannot capture. Dessa was built for real-time conversation from the start and trained only on human speech audio.

What remains open is access. The model and API are not public, and the team is still recruiting AI safety researchers to examine risks, so any broader availability appears to depend on that review rather than a stated date.

FAQ
Can I try Dessa now?
No.
What hardware does Dessa run on?
It was trained on a single NVIDIA GB200 NVL72 and is said to run locally on an M4 Max MacBook Pro.
How is Dessa different from typical voice assistants?
Typical voice assistants wait for a speaker to finish before responding, while Dessa can reply over a speaker, backchannel and laugh in reaction.
Send on LINE

Also reported by The Verge AI

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleMicrosoft: AI speeds attacks, but entry points unchanged