Block introduced Dessa on Oct. 23, 2026, presenting a speech model trained on human conversation rather than text. The announcement expands the company’s AI research into voice interaction, while leaving public availability and commercial deployment unresolved.
This Rapid News Brief covers a research announcement, not a product that businesses can already install. Block’s public page invites a small group of AI safety researchers to seek early access and says launch links are coming soon. TechCrunch reported the model on Oct. 24.
A different route to voice interaction
Block describes Dessa as a full-duplex system: it can receive speech while speaking, instead of requiring each side to wait for a completed turn. The company says training used human speech without text or pretrained models. It also says inference can run locally on a MacBook Pro with an M4 Max chip or better.
Those are the developer’s claims, not independent benchmark findings. Block has published demonstration conversations, but demonstrations alone do not establish accuracy, reliability or suitability for a business workflow.
Research access comes before deployment
The research group is led by Ragavan Thurairatnam and reports to Block co-founder Jack Dorsey, according to the announcement. Block says outside safety researchers are being invited to investigate risks the team may not have anticipated.
For product and customer-experience leaders, the relevant distinction is between conversational behavior and an operating service. A voice system can sound responsive without having established the information accuracy, access controls or escalation process needed for a particular task. Those are evaluation questions, not findings about Dessa’s performance.
Local execution is also a technical claim, not proof of a deployable package. A hardware description does not specify release terms, support, integration requirements or the cost of running a commercial service.
The announcement therefore provides a research direction to watch rather than a purchasing decision. Block’s early-access invitation makes further testing part of the next stage; public demonstrations should not be treated as evidence that the model is ready to handle customer interactions without additional evaluation.