Block Unveils Dessa, Speech Model It Says Learned From Conversation
Block's in-house AI research lab announced Dessa on 23 October 2026. Dessa is a speech-to-speech model that listens and talks at the same time, and Block says it was trained from scratch on human conversation, without text or pretrained models. It is a research preview with a public waitlist and early access for a small group of AI safety researchers. No price, API or public release date has been announced. For business owners, Dessa is not something to buy or deploy yet. It is a signal that voice-based automation may soon skip the text step that most phone and chat tools rely on today.
In short
- Block announced Dessa, a full-duplex speech-to-speech model, on 23 October 2026 as a research preview.
- Block says Dessa was trained only on human conversation audio, with no text and no pretrained models.
- Access is limited to a public waitlist and a small group of AI safety researchers, with no price or release date.
- Dessa is not a product a business can buy today, so voice workflows should be mapped now and tested later.
What happened
Block's in-house AI research lab announced Dessa on 23 October 2026, according to Block's own announcement and coverage by Unite.AI. Dessa takes audio in and gives audio out. It is full-duplex, which means it can listen while it talks, and Block says it learned from real conversations when to cut in and when to offer a short acknowledgement such as "yeah".
Block says Dessa was trained from scratch on human speech, with no text and no pretrained models. According to Unite.AI, the company also says the model runs locally on a MacBook Pro M4 Max or better. The reports differ on the training hardware. TechCrunch describes a single NVIDIA GPU rack, while Voice AI Space reports a single NVIDIA GB200 NVL72 system.
Access is limited. According to Digg, Block offers a waitlist, and early access goes to a small group of AI safety researchers. Voice AI Space reports that no API, pricing or model weights have been announced.
Why it matters for businesses that automate work
Most automated phone and chat tools first turn speech into text, then run a language model on that text, then turn the reply back into speech. Voice AI Space notes that the transcript step can drop signals such as emotion, sarcasm and laughter. Dessa is built to skip that step, working from audio to audio.
If the approach holds up outside Block's demos, it could change how voice intake works for customer calls, order changes and simple payment queries. Block's claims are its own. TechCrunch reports that the company's researchers posted examples of conversations, and those examples are the main public evidence so far.
What changes in practice
For now, nothing changes in the tools you buy. Dessa is a research preview, and a business cannot sign up for it. Your voice workflows will still run on the speech-to-text stack you already use, or on the one your vendor chose.
There are three practical steps for an owner or operations head:
- List every workflow where a person currently answers calls or chats to do a routine task, such as order status, rescheduling or payment reminders.
- Record how many of those calls end in a simple action in your billing, CRM or ticketing system. These are the calls a voice agent could handle first.
- Ask any voice vendor whether their model is in production, which languages it supports, and how call recordings and customer consent are handled in each country where you operate.
Dessa may change the quality of voice automation, but it will not remove the need for checks. A voice agent that sounds natural can still book the wrong date or quote the wrong amount. A person still needs to review exceptions, complaints and any call that touches money or contracts.
What to watch next
Four signals will show whether Dessa moves from demo to product. The first is whether the waitlist opens to developers and businesses. The second is whether Block publishes pricing, a release date or model weights. The third is independent testing, since Block's page does not include a published benchmark. The fourth is whether Block shares latency and accuracy figures that other teams can check.
Until those arrive, the useful action is preparation. Map your voice-driven routine work now, so that a pilot can start quickly when a tool is ready to test.
How AiStaffo would automate this
AiStaffo would start by listing the calls and chats your team handles for routine tasks, such as order status, rescheduling and payment reminders, and connecting the channels where they arrive to your billing, CRM or ticketing system. Routine requests that match a clear rule would be logged and actioned automatically, with a full record in your system. Exceptions, complaints and anything involving money or contracts would be passed to a named person with the call summary attached. You would still approve the rules, the scripts and the escalation list. Book a free automation audit.
Questions people ask
What is Block's Dessa?
Can my business use Dessa today?
How is Dessa different from a typical voice assistant?
Does Dessa run on a normal office computer?
Should my business change its voice automation plans because of Dessa?
Book a free automation audit
Thirty minutes. We look at one process you run every week and tell you exactly what an AI worker would take off your desk, and what it would not.
