Block unveils Dessa voice AI that laughs and runs on a MacBook

Block, the US fintech led by co-founder and CEO Jack Dorsey, unveiled its voice AI Dessa on October 23, 2026, according to TechCrunch. The model was developed by the company's own research organization and is designed to hold a natural spoken conversation, laughing, pausing and shifting tone with the emotional flow of the exchange.

Dessa uses full-duplex technology, meaning it can listen to the other speaker and respond at the same time rather than waiting for the user to finish. Block said the model was trained only on real human conversation audio, without converting speech to text and without building on another pretrained AI model.

The company said it trained Dessa on a single rack of Nvidia GPU servers, and that the model can run on a MacBook Pro with an Apple M4 Max chip or a more powerful computer, without large data centers.

Dessa is not yet available to general users. Block said it will give AI safety researchers early access to check for risks, and has not disclosed a schedule for applying the model to its payment services such as Square and Cash App. The conversation videos released so far are demonstration materials produced by Block, and no independent performance verification has been published.

Dorsey said voice is the most natural way to interact with intelligence and that talking with Dessa feels like talking to a person. OpenAI previously released its GPT-Live voice model, which also supports simultaneous two-way conversation, and Square already uses voice AI to handle food orders. Block shortened its name from Square in 2021.