Block Dessa is a speech-to-speech AI model from Block, and my verdict is simple, which is that it looks like the most natural-sounding voice research I've seen, but it isn't a product you can use yet.
Block, the company behind Square and Cash App, built it in a very small research lab that reports directly to Jack Dorsey.
It was trained from scratch on human conversation audio, with no text and no pretrained models, and it listens and talks at the same time.
Block's page says it runs locally on a MacBook Pro M4 Max or better, and on 24 October 2026 the launch links were marked as coming soon.
I haven't tested Dessa myself, so this comparison is built on what Block has published, set against the voice setups I run every day.
That's the honest way to judge a launch like this, because the only hands-on evidence anyone has is Block's own demo recordings.
I don't have a Dessa video, so here's my video on Hermes voice agents, which is the kind of voice stack I'm comparing it against.
🔥 Want my exact Block Dessa vs Today's Voice 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
The Block Dessa Verdict At A Glance
These are my scores for the launch, based only on what Block and its team have published.
| Category | Score | Why it gets that score |
|---|---|---|
| Conversation quality shown | 9/10 | The demos show laughing mid-sentence, interrupting, finishing a forgotten title and reacting to a surprise. |
| Technical ambition | 9/10 | Training a conversational model on speech alone, with no text, is a genuinely different approach. |
| Team credibility | 8/10 | Block says the team has built speech models for over eight years and deep learning models since 2012. |
| Hardware needs | 8/10 | It runs locally on a MacBook Pro M4 Max or better, which is expensive but still a laptop. |
| Openness about safety | 7/10 | Block is inviting outside safety researchers, but it hasn't published findings yet. |
| Evidence you can check | 5/10 | The only evidence is Block's own recordings, and nobody outside Block has tested it publicly. |
| Developer readiness | 2/10 | The page lists no API, no SDK and no docs. |
| Availability | 2/10 | The download and waitlist links were marked as coming soon, with no date or price. |
My overall verdict is that Block Dessa wins the research round and loses the product round, for now.
What Block Actually Launched
Block Dessa is a full-duplex speech-to-speech model, which means audio goes in, audio comes out, and it can talk and listen at once.
Block says it was trained from scratch on human conversation, with no text and no pretrained models.
The page says it natively hears and speaks emotion, sarcasm, laughter, a frustrated sigh and a voice cracking with sadness.
It also says inference runs faster than real time, and that Block trained the model on only one NVIDIA GB200 NVL72.
Ragavan Thurairatnam, who signs the page as Dessa lead, called it a "GPT3 moment for speech" on X.
That's a bold claim from the person who built it, so I'm treating it as a goal to test rather than a result.
Round 1: Block Dessa vs A Cascaded Voice Agent
Most voice AI, including mine, is cascaded, which means a chain of separate tools passes your words along.
| Area | Cascaded voice agent | Block Dessa | Winner |
|---|---|---|---|
| Hearing you | Speech becomes text before the model sees it. | Speech goes straight into the model. | Dessa wins, because nothing is lost in translation. |
| Turn-taking | It usually waits for you to stop. | It listens and talks at the same time. | Dessa wins on Block's demos. |
| Emotion | Tone is mostly lost in the text step. | Block says it hears and speaks emotion natively. | Dessa wins on Block's description. |
| Speed | Every hand-off adds a little delay. | Block says inference runs faster than real time. | Dessa wins on paper, but I haven't timed it. |
| Doing tasks | It can call tools, open apps and run jobs. | The page doesn't mention tools. | The cascaded agent wins clearly. |
| Availability | You can build one today. | The launch links were marked as coming soon. | The cascaded agent wins today. |
Block's argument is that the text step throws away information text alone can't carry.
Its page uses the example of one sentence said with the stress on different words, where the meaning changes even though the words don't.
That's a fair hit on every cascaded system, including the ones I use.
Round 2: Block Dessa vs My Own Voice Agents
My Agent OS has a voice copilot called Apollo, and I also use Hermes Jarvis, so this is the comparison I care about most.
Apollo's loop is a wake word, then speech to text, then a brain that answers or hands the job to an agent, then a butler voice that speaks back.
Apollo also has a realtime mode, where I talk continuously and can interrupt it mid-sentence.
| Area | Apollo and Hermes Jarvis | Block Dessa |
|---|---|---|
| What it's for | It runs my Mac by voice, opens apps and builds apps while I watch. | Block presents it as a natural conversation model. |
| Memory | Apollo writes what I tell it to my Obsidian vault and reads me a briefing. | The page doesn't mention memory. |
| Emotion | It hears my words, but not really my tone. | Block says it hears tone natively. |
| Where it runs | It runs in my Agent OS on my own Mac. | It runs on a MacBook Pro M4 Max or better. |
| Can I use it today? | I use it every day. | The launch links were marked as coming soon. |
My verdict for this round is that my agents win on getting work done, and Dessa looks set to win on how the conversation feels.
The dream setup would put a Dessa-style voice in front of an agent like Hermes, but Block hasn't published a way to connect the two.
If you want to see the agent side of that setup, read my post on the free Hermes Agent OS.
🔥 Want the voice agent setup that wins Round 2 today? Inside the AI Profit Boardroom, I've got the Agent OS with Apollo and my Hermes voice setup, plus step-by-step tutorials. You'll be learning alongside 3,400+ members with four coaching calls every week. → Get access here
Round 3: Block Dessa And GPT-Live-1
Block includes a clip of Dessa talking with GPT-Live-1, which it identifies as ChatGPT Live.
Both voices in that clip are AI-generated, and a second sample called Noisy GPT live appears in the extra recordings.
Block is careful here, and it says the developer of GPT-Live-1 isn't affiliated with Block and hasn't reviewed or endorsed Dessa.
Block doesn't publish a scored comparison between the two, so I'm not going to declare a winner.
What the clip does show is Block's confidence, because putting your model in a live conversation with a rival's is a bold choice.
Round 4: The Team And The Evidence
The team behind a launch is part of the verdict, so here's what Block says about it.
Block describes a very small AI research lab, founded by former members of Dessa, the Toronto deep learning startup.
The lab says it has built deep learning models since 2012 and generative speech models for over eight years.
It reports directly to Jack Dorsey, who the page lists as Block's cofounder and chairman.
The evidence side is weaker, because every demo is Block's own recording.
Block says nobody steers Dessa in those clips and that subtitles were added afterwards, which is useful transparency.
Still, I'd want independent testers and safety researchers to report back before I'd call anything proven.
The Numbers Block Published
Block says Dessa learned language from hundreds of times fewer words than large language models need.
The page says a single spoken word takes thousands of times more bytes than the same word written as text.
It also says speech never repeats exactly, so the same sentence said twice still gives different data.
The page doesn't publish model size, hours of training audio, supported languages or benchmark scores.
If you want a second opinion on which AI tools are worth your budget right now, you can book a free SEO strategy session with my team.
Who Should Care About Block Dessa
Different readers should take different things from this launch.
- Voice AI builders should study the demos, because they show how far natural turn-taking has moved.
- AI safety researchers can apply for early access through the form on Block's page.
- Business owners should keep using agents for real work, and test Dessa once the launch links go live.
- Anyone with a MacBook Pro M4 Max or better should bookmark the page, because that's the hardware Block names.
For a full walkthrough of the product itself, read my feature-by-feature post on Dessa AI.
For another recent AI launch I've broken down, read my post on the Hark AI agent.
🔥 Want my honest verdicts on every new AI tool? The AI Profit Boardroom has the Agent OS, AI tool training and four weekly coaching calls where you can ask me which launches are worth your time. → Join the AI Profit Boardroom
Related Reading
Also On Our Network
- 🌐 Seven things operators should know about Block Dessa
- 🌐 Why a payments giant built Block Dessa, and what it means for business
- 🌐 Block Dessa explained for Agent OS users
- 🌐 How Block Dessa learned to talk without any text
FAQ: Block Dessa Verdict
Is Block Dessa better than other voice AI?
On how natural it sounds in Block's demos it looks very strong, but on usefulness today it loses because there's no download, price or API yet.
How is Block Dessa different from a cascaded voice agent?
A cascaded agent goes through text, while Block Dessa takes audio in and gives audio out.
Did Block compare Dessa with ChatGPT?
Block shows a clip of Dessa talking with GPT-Live-1, but it doesn't publish a scored comparison.
Who is behind Block Dessa?
A very small Block lab founded by former members of the Toronto startup Dessa built it, and it reports directly to Jack Dorsey.
Should I wait for Block Dessa or use a voice agent now?
Use a voice agent now for anything that needs actions, and test Dessa when the launch links go live.
If you'd like my team to help you choose the right AI setup for your business, book a free SEO strategy session and we'll go through your options.
📺 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
My verdict holds until I can test it myself, which is that the voice research is exceptional and the product is still on its way with Block Dessa.
