Back to Blog

AI Model Watch · Research Preview

Dessa AI: What It Is, Demos & Access Status

Dessa AI is a conversational voice model from Block. According to the company, it can listen while speaking, allowing interruptions and short replies during a conversation. You can hear examples on the official page, though public access was still limited when we checked on October 25, 2026.

By Froging AI Editorial Team · · 6 min read

Original editorial illustration of amber and green sound ribbons overlapping between two sculptural listening forms
AI-generated illustration for this article, showing overlapping sound waves. It is not a Dessa screenshot or output.

What is Dessa AI, and who makes it?

Dessa is built by a small research team at Block, the company behind Square and Cash App. According to its introduction, the team was founded by former members of the Toronto startup Dessa and reports directly to Jack Dorsey.

Block says it trained Dessa from scratch using only recordings of human conversation, without text or pretrained models. Read Block’s introduction

Where to hear the official Dessa demos

The official page has recordings of people talking to Dessa. Block says the model chooses its replies, timing and voice during each conversation; subtitles were added afterward. Hear the official recordings

Selected Dessa recordings and Block’s descriptions
Recording on the official pageWhat Block highlights
Talking about the releaseLaughter during a sentence and a louder response when asked to speak up.
Remembering a movieHelping the speaker recall a movie title by completing their thought.
An anniversary surpriseA gasp and a change in tone in response to a surprise.
Expressive conversationChanges in speaking pace and intonation.

These descriptions come from Block’s demo notes. Listen to several recordings rather than judging from one short clip. A natural voice does not guarantee a correct answer.

Dessa AI access, waitlist, download and pricing

The research-access form asks for your name and institutional email. You can also include a Google Scholar or blog link. Block decides who gets access.

For general users, the download and waitlist buttons were not yet connected to release links when we checked. Return to the official page for updates.

The page did not list prices, public API documentation, downloadable model weights or a software license. There are no confirmed plans or installation steps to share yet.

Block says Dessa runs locally on a MacBook Pro M4 Max or better. It has not provided a full installation guide, so memory requirements, supported operating systems and licensing still need clarification.

What does full-duplex voice mean?

Full-duplex means listening and speaking can happen at the same time. Think about saying “yes” while someone finishes a sentence, or interrupting to correct a detail. A conversation system needs to handle those overlapping signals rather than treat every sound as a completely separate turn.

Separated turns

Person speaks → finishes
AI responds → finishes

Overlapping conversation

Person speaks ↔ can interrupt
AI listens ↔ can respond
A simple illustration of turn-taking, not a timing measurement or a diagram of the model.

Block describes Dessa as a speech-to-speech model: it receives audio and produces audio. The aim is to retain details such as tone, laughter and pauses that a plain transcript does not capture.

What the preview still leaves unanswered

The demos leave several practical questions unanswered: how well does Dessa handle background noise, unfamiliar topics or people talking over it? Which languages does it support? The announcement does not include enough test results to answer those questions.

Is Dessa available on Froging AI?

No. Froging AI does not offer Dessa or distribute invitations. Use Block’s page for access updates.

Sources and update notes

Primary source: Block — Dessa announcement, recordings and research access invitation. We checked the page on October 25, 2026. The training and hardware details in this article come from Block. We have not installed or independently tested the model.

For changes to access, downloads, prices or API availability after that date, check Block’s page.

Dessa AI FAQ

Quick answers about the official website, demos and release status as of October 25, 2026.

What is Dessa AI?

Dessa is a conversational speech model developed by a research lab at Block. Block says it was trained from scratch on human conversation audio and can listen and speak at the same time.

Can I try Dessa AI right now?

On October 25, 2026, you could listen to recordings or apply for research access. We found no working public trial, download or waitlist signup on the official page.

Where is the official Dessa AI website?

Visit Block’s Dessa page using the link above. It has the recordings and the research-access application.

Is Dessa AI free, or does it have paid plans?

Block had not published prices on the official page as of October 25, 2026. It is too early to say whether a public release will be free or paid.

Can I download Dessa or run it on a Mac?

Block says Dessa runs on a MacBook Pro M4 Max or better, but there was no working download link when we checked on October 25, 2026. A full installation guide was not available.

Is Dessa open source, and does it have an API?

As of October 25, 2026, the official page did not provide public API documentation, model weights or a software license. Being able to run locally does not mean a model is open source.

Can I use Dessa on Froging AI?

No. Froging AI makes images and videos. It does not offer Dessa or live voice conversations.