Dessa AI is a full-duplex speech-to-speech model from Block, and it was trained from scratch on human conversation audio with no text and no pretrained models.
Block, the company behind Square and Cash App, built it in a small research lab that reports directly to Jack Dorsey.
Block's own page says it runs locally on a MacBook Pro M4 Max or better, and that inference runs faster than real time.
When I checked the official page on 24 October 2026, it showed a Download button, a Join the waitlist button and a note that launch links are coming soon.
I haven't used Dessa myself yet, so everything below comes from Block's page at block.xyz/dessa and the posts its team shared on X.
This page is the operator walkthrough, which means what it is, what you need, how to get in and what to test on day one.
I don't have a Dessa video, so here's my video on a Hermes voice agent, which shows the kind of voice setup Dessa is trying to beat.
Dessa AI At A Glance
This table is the whole product on one screen, so you know exactly what you're dealing with before you sign up for anything.
| Question | What Block's page says |
|---|---|
| What is it? | It is a full-duplex speech-to-speech model, so audio goes in and audio comes out. |
| Who made it? | A very small AI research lab at Block made it, and the lab reports directly to Jack Dorsey. |
| What was it trained on? | It was trained from scratch on human conversation audio, with no text and no pretrained models. |
| What was it trained with? | Block says it trained Dessa on only one NVIDIA GB200 NVL72. |
| Where does it run? | It runs locally on a MacBook Pro M4 Max or better, according to the page. |
| How fast is it? | Block says inference runs faster than real time, so there's no awkward wait. |
| Can I get it today? | The page shows Download and Join the waitlist buttons and says launch links are coming soon. |
| What does it cost? | The page doesn't list a price, a plan or an API. |
That last row matters, because you shouldn't plan a client project around a model you can't install yet.
🔥 Want my exact Dessa AI setup, step by step? Inside the AI Profit Boardroom you get step-by-step video trainings, my installable Agent OS and four coaching calls a week with 3,400+ members building real automations. → Join the AI Profit Boardroom
What Dessa AI Actually Is
Dessa AI is a conversation model that learned to talk by listening, the way a child does.
Block's page makes the comparison directly, which is that babies learn to talk without ever seeing a word written down.
The lab wanted to see whether a model could do the same, and its answer is that it can, but that it's very hard.
The model is full-duplex, which means it listens and talks at the same time instead of waiting for you to finish.
That's why it can cut in at the right moment and drop in little backchannel sounds that show it's following you.
It also hears and speaks emotion natively, so sarcasm, laughter, a frustrated sigh or a voice cracking with sadness all reach the model as sound.
Block's claim is that, as far as it knows, Dessa is the first AI you can have a realistic live conversation with that was trained only from human speech.
I'd treat that as Block's claim rather than a settled fact, because nobody outside Block has tested it publicly yet.
Why Dessa AI Works Differently From The Voice Agents You've Used
Most voice AI you've used is a chain of three separate pieces.
The first piece turns your speech into text, the second piece is a language model that writes a reply, and the third piece reads that reply out loud.
Block's argument is that the first step throws away a huge amount of information that text can't carry.
The page says the most powerful AIs in the world can't hear how you feel, and the lab calls that a worse experience and a potential danger.
Here's how the two approaches compare in plain terms.
| Step | A typical voice agent | Dessa AI |
|---|---|---|
| Input | Your voice is turned into text before the model sees it. | Your voice goes straight into the model as audio. |
| Turn-taking | The agent usually waits for you to stop talking. | The model listens and talks at the same time. |
| Emotion | Tone, sighs and laughter are mostly lost in the text step. | Block says the model hears and speaks emotion natively. |
| Output | A separate voice engine reads the text reply aloud. | The model produces the speech itself, including its own voice. |
| Speed | Each hand-off adds a little delay. | Block says inference runs faster than real time. |
My own voice agents work the first way, and they're brilliant at doing jobs, but they don't hear tone.
That's the gap Dessa is aiming at.
What You Need To Run Dessa AI
The only hardware Block's page names is a MacBook Pro M4 Max or better.
Ragavan Thurairatnam, who signs the page as Dessa lead, went a step further on X and described it as a "full-duplex speech-to-speech model that you can run on consumer GPUs".
Jack Dorsey also posted on X that it's small enough to run on a MacBook.
So the honest operator checklist today looks like this.
- You check whether you own a MacBook Pro with an M4 Max chip or something more powerful.
- You keep an eye on Block's page, because the Download button is the route the page shows for getting the model.
- You join the waitlist when the link goes live, because that's the other route the page shows.
- You don't buy new hardware yet, because Block hasn't published full system requirements, memory needs or supported operating systems.
I'd rather tell you what isn't known than guess at it.
The page doesn't say how big the model is, which languages it speaks or whether there will be a cloud version.
How To Get Access To Dessa AI
There are three doors on Block's page, and only one of them is fully open right now.
Door 1: Listen to the demo recordings
This door is open today, and it's the best place to start.
Every clip on the page is a live recording of a person talking to Dessa, and Block says nobody steers it.
Dessa chooses what to say, when to say it and how to say it, right down to its own voice and personality.
Block also says the subtitles were added afterwards, so they aren't part of what the model produced.
Each clip plays as two synchronised speaker tracks, which lets you hear both sides of the conversation at once.
Door 2: The Download and waitlist buttons
The page shows a Download button and a Join the waitlist button at the top.
On 24 October 2026 the page also said launch links are coming soon, so I'd bookmark it and check back.
Door 3: Early access for safety researchers
Block is giving a small group of AI safety researchers early access to find risks the lab hasn't anticipated.
The form asks for your name, an institutional email from a university, company or lab, and an optional Google Scholar or blog link.
If you're not a safety researcher, don't apply through this door, because it's clearly meant for people who test models for risk.
🔥 Want a voice agent that does real work while you wait for Dessa AI? Inside the AI Profit Boardroom, I've got the Agent OS with my Hermes voice setup, plus step-by-step video tutorials on wiring voice into your agents. You'll be learning alongside 3,400+ members with four coaching calls every week. → Get access here
Every Behaviour Block Shows In The Dessa AI Demos
This is the part I'd study first, because it tells you what to test when you finally get your hands on it.
| Demo clip | What Block says happens |
|---|---|
| Talking about the release | Dessa laughs the moment something is funny, mid-sentence, and it gets louder when asked to speak up. |
| Remembering a movie | The person forgets a film title and Dessa finishes the sentence for them. |
| An anniversary surprise | Dessa gasps under its breath and answers with what sounds like teary gratitude. |
| Expressive conversation | The clip shows a huge range in intonation and speed. |
| Life advice | Block says Dessa sounds genuinely empathetic here. |
| Dessa in conversation with GPT-Live-1 | Dessa talks to ChatGPT Live, and both voices in the clip are AI-generated. |
Block adds a clear note on that last clip, which is that the developer of GPT-Live-1 isn't affiliated with Block and hasn't reviewed or endorsed Dessa.
Below those six, the page lists 15 more samples with names like Refund, Safety lab, Zelda discussions and Singing nonsense Spanish.
Block also says more recordings will be shared on X soon.
The two behaviours I'd watch closely are the laugh and the volume change, because Block says nobody taught it to get louder.
The Dessa AI Facts That Surprised Me
A few numbers on the page are worth knowing, because they explain why this is a big deal for people who build agents.
Block says Dessa learned language from hundreds of times fewer words than large language models need.
The lab is fair about it, and it jokes that language models also have to learn programming languages that Dessa doesn't.
Block also says a single spoken word takes thousands of times more bytes than the same word written down.
Speech never repeats exactly, so even the same sentence said twice by the same person comes out as different data.
Despite that, Block says it trained Dessa on only one NVIDIA GB200 NVL72.
The lab also says Dessa knows a surprising amount of facts, considering every fact came from raw sound.
What I'd Test First When Dessa AI Opens
Once the download goes live, here's the order I'd run my first session in.
- I'd interrupt it mid-sentence to see how it handles being cut off.
- I'd ask it to speak up and then to speak quietly, because that's one of Block's headline behaviours.
- I'd tell it a joke with a bad punchline to see whether it laughs or groans.
- I'd forget a word on purpose to see whether it fills the gap.
- I'd sound frustrated on purpose to see whether its tone changes.
- I'd time how quickly it replies on the MacBook Pro the page names.
That gives you a real picture in about 15 minutes, which is better than reading another hype thread.
Dessa AI Limits And Open Questions
Here's what I'd want answered before I put Dessa anywhere near a client.
| Open question | Why it matters to an operator |
|---|---|
| The page shows no release date for the download. | You can't plan a launch around a model with no date. |
| The page shows no price, plan or API. | You can't quote a client until you know the cost. |
| The page doesn't say which languages it speaks. | Block lists new languages as one of its own research questions. |
| The page doesn't mention tools or actions. | A conversation model isn't the same as an agent that opens apps or sends emails. |
| The safety work is still early. | Block is inviting outside researchers precisely because the risks aren't fully mapped. |
The safety angle is a real part of Block's story, not a footnote.
The page says AI is getting more powerful and independent while staying nearly deaf to emotion in our voices.
Block's example is someone in serious pain while a voice AI has no idea and just keeps going.
The lab says Dessa is a step toward fixing that.
If you want help working out where voice AI fits your business, you can book a free SEO strategy session with my team.
Where Dessa AI Fits In An Operator's Stack
I'd think of Dessa as a voice, not as a whole agent.
My Hermes voice agents open apps, build pages and remember things, because they sit on top of a full agent with tools.
Dessa, as Block describes it, is about how natural the conversation feels.
The interesting future is putting the two together, so the agent does the work and a speech model like Dessa handles the talking.
Until Block publishes how developers can connect to it, that's a hope rather than a setup guide.
If you're building agent workflows today, my guide to the free Hermes Agent OS and my list of the best Hermes Agent use cases are the places to start.
For the company and news side of this launch, read my post on Block Dessa, which covers the team and why Block built it.
🔥 Want my voice and agent setup done with you? The AI Profit Boardroom has the Agent OS, Hermes Agent training and four weekly coaching calls where you can ask me how to add voice to your agents. → Join the AI Profit Boardroom
Related Reading
- Block Dessa, the company and the news behind the model
- The free Hermes Agent OS and what's inside it
- The best Hermes Agent use cases, ranked
Also On Our Network
- 🌐 Six ways a business could use Dessa AI once it opens
- 🌐 Dessa AI reviewed feature by feature, with early scores
- 🌐 The Dessa AI quick start for an Agent OS
- 🌐 What GoldieBench measures, and why Dessa AI isn't on it
FAQ: Dessa AI
What is Dessa AI?
Dessa AI is a full-duplex speech-to-speech model from a small research lab at Block, and Block says it was trained only on human conversation audio.
Can I download Dessa AI today?
The official page showed Download and Join the waitlist buttons on 24 October 2026, along with a note that launch links are coming soon.
What hardware does Dessa AI need?
Block's page says it runs locally on a MacBook Pro M4 Max or better.
How much does Dessa AI cost?
Block's page doesn't list a price, a plan or an API, so I'd wait for the launch links before budgeting.
How is Dessa AI different from normal voice AI?
Normal voice AI turns speech into text first, while Dessa takes audio in and gives audio out, so Block says it keeps emotion and timing.
Can safety researchers get early access to Dessa AI?
Yes, Block says a small group of AI safety researchers is getting early access, and the form asks for an institutional email.
If you'd like my team to look at how voice agents could fit your business, book a free SEO strategy session and we'll talk it through.
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
About Julian
I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
- I've built a 7-figure SEO and link-building agency, Goldie Agency, with 70+ people in Bangkok.
- I've grown a YouTube channel to 400,000+ subscribers.
- I wrote the book "Link Building Mastery", which covers nine link building strategies.
- I run GoldieBench, where I score AI models on real one-shot builds.
→ Get my best AI training inside the AI Profit Boardroom
Listen to the demos today, check the page for the launch links and run those six tests on day one, and you'll know quickly whether Dessa AI is worth building around.
