Block's AI research lab unveiled Dessa, a speech-to-speech model it says was trained from scratch on nothing but human conversation — no text, no pretrained models. It listens and talks at once, handles emotion and laughter, and runs locally on a high-end Mac.
2026-10-23 · by Moe Ameen
Block — the payments company co-founded by Jack Dorsey — unveiled Dessa in October 2026 through its small in-house AI research lab, which reports directly to Dorsey and was founded by former members of Dessa, the Toronto deep-learning startup Square (now Block) acquired in 2020. The model carries that name. Dessa is a full-duplex speech-to-speech conversation model: audio in, audio out. Block describes it as "the most authentic full-duplex speech model ever built" and says that, as far as the team knows, it is the first AI you can hold a realistic live conversation with that was trained only from human speech. Those are the company's own claims; no independent benchmarks were available at announcement.
The notable part is how it was trained. Block says Dessa learned entirely from human conversation — "sounds, no text, no pretrained models" — on a single NVIDIA GB200 NVL72 system, and that it picked up language from hundreds of times fewer words than large language models. Because it is full-duplex, it listens and talks at the same time, and it learned from real conversation when to cut in and when to backchannel with short acknowledgments like "yeah" or "got it." Block says the model natively hears and speaks emotion, sarcasm, laughter, a frustrated sigh, and a voice cracking with sadness, rather than flattening speech into text first. Inference, the company says, runs faster than real time, so replies come without an awkward pause.
On availability, Dessa is early. Block opened a public waitlist and is offering early access to a small group of AI safety researchers through an application that asks for an institutional email; it says launch links are coming soon. The model reportedly runs locally on a MacBook Pro M4 Max or better, and lead researcher Ragavan Thurairatnam said it can also run on consumer GPUs. Block has not announced pricing, a general release date, or an open-source license. The public demos include a recorded conversation between Dessa and GPT-Live-1, a third-party model whose developer Block says is not affiliated with it and has not endorsed Dessa.
You can't publish a live conversation with Dessa today — it's waitlist-only — but the shift it represents is worth acting on now, because the conversations you already record are a content goldmine that mostly goes to waste. That's the gap [Kompozy](/) closes. Record a podcast episode, an interview, or a long voice memo, and Kompozy turns the spoken source into a full week of posts: vertical [Clipped Shorts](/glossary/clipped-short) cut from the best moments and auto-captioned for TikTok, Reels, and Shorts; Quote Graphics pulled from the strongest lines; a Blog Article and an Email Newsletter written from the transcript; and Text Posts for X and LinkedIn — every piece governed by your [Persona Brief](/glossary/persona-brief) so one voice and one banned-word list hold across all of them.
Where Dessa hints at natural spoken AI, Kompozy already ships the branded version of that for social: face-locked [Persona Shorts](/glossary/persona-shorts) present a recurring AI-influencer avatar with its own built-in voice, and [HyperFrames](/glossary/hyperframes) renders carousels and graphics pixel-exact to your brand. [Autopilot](/glossary/autopilot) then schedules and fans the whole set across the eight social platforms plus blog and email behind a per-post review gate. Dessa is pushing on how machines talk; Kompozy is the engine that turns talk — yours or a guest's — into a published, on-brand presence everywhere your audience is.
Dessa is a full-duplex speech-to-speech conversation model from Block's in-house AI research lab, unveiled in October 2026. Block says it was trained from scratch on human conversation alone — no text and no pretrained models — so it listens and talks at the same time, backchannels, and handles emotion, laughter, and sighs. It is named after Dessa, the Toronto deep-learning startup Square (now Block) acquired in 2020.
Not generally yet. At announcement Block opened a public waitlist and offered early access to a small group of AI safety researchers via an application requiring an institutional email, saying launch links are coming soon. It reportedly runs locally on a MacBook Pro M4 Max or better. Block has not announced pricing, a general release date, or an open-source license.
Two claims stand out. First, it was trained only on human speech, not text, which Block says lets it carry emotion, sarcasm, and timing more naturally. Second, it is full-duplex and runs locally, so it can listen and talk at once with faster-than-real-time inference on a high-end Mac. These are the company's own claims; independent benchmarks were not available at announcement.
You don't need Dessa to turn conversation into content. Record a podcast, interview, or voice note and a content engine like Kompozy repurposes the spoken source into captioned short clips, quote graphics, a blog, a newsletter, and platform-specific text posts, then schedules and publishes them across the eight social platforms plus blog and email.