Jack Dorsey’s company Block announced on October 23, 2026 that it has developed a new voice model named Dessa, created in the company’s research lab. The model is described as full‑duplex — able to speak and listen simultaneously — and the team published multiple demo examples on the day of the announcement.
Capabilities and technical approach
Block says Dessa was trained entirely on human speech and can run on a single NVIDIA GPU. The company emphasizes that the model produces human‑like expressive behavior such as laughs, pauses, crying‑like sounds, and a natural cadence. According to Block, this emotional nuance is built into the model itself rather than being driven by text‑in‑the‑loop processing or external “voice tags,” which are used by other systems.
Full‑duplex here means the model can talk and listen at the same time, a capability similar to the GPT‑Live model OpenAI released earlier in the year. Block also stated that Dessa’s knowledge comes from conversational data it has encountered, rather than relying on a separate large language model for retrieval.
Demonstrations shared by the team
Researchers from Block posted several conversation examples showing Dessa switching voices, cadences and emotional tones. In one clip the agent says it is tired from work and the voice sounds tired; in another example the model responds emotionally and nearly breaks down. A separate demo shows a Block team member playing a guessing game with the assistant, where the model quickly guesses correctly.
Availability and safety testing
Block said Dessa can run locally on consumer‑grade computers such as a MacBook M4 Max or better. However, the model is not publicly accessible at this time. The company has invited safety researchers to test Dessa, likely to explore scenarios where a voice agent could deceive people on the other end of a conversation.
Business implications
Block is primarily a financial technology company that sells solutions to small businesses and already uses voice AI in some Square ordering flows. Having an on‑device, emotionally expressive voice model could enable more business and consumer experiences at lower cost and with reduced dependency on cloud processing, according to the company.
