Block Releases Dessa, a Full-Duplex Voice Model That Runs Locally on a MacBook
Jack Dorsey's Block has released Dessa, a full-duplex voice model from its research lab that can speak and listen simultaneously. Trained entirely on human speech, it runs locally on a single NVIDIA GPU and consumer hardware such as a MacBook M4 Max, and simulates emotional expressions including laughter, pauses, and crying. Block says its knowledge comes from conversational data without pairing with a large language model. Dessa is not publicly available; Block has invited safety researchers to test it.
IllustrationEditorial responsibility
- No named human review is recorded for this page.
- Reports are grouped by semantic similarity and deterministic rules. Language models may assist titles, summaries, translation and cross-source analysis; the page reads the event directly, while its address stays stable when the title changes.
- Summary covers the current reports
Cross-source coverage
Reporting timeline
Block Releases Full-Duplex Voice Model Dessa for Local MacBook Use
Jack Dorsey's Block has launched Dessa, a full-duplex voice model that can speak while listening to user speech. According to the report, Dessa can simulate emotional expressions such as laughter, pauses, and choking up. It is trained entirely on human speech data, and its knowledge comes from training dialogue data without needing to be paired with a large language model. The model is said to support local operation on M4 Max and above MacBooks.
Read sourceBlock Releases Dessa, a Full-Duplex Voice Model That Sounds Eerily Human
Jack Dorsey's Block unveiled Dessa, a new full-duplex voice model from its research lab that runs locally on consumer hardware such as a MacBook M4 Max. The company says Dessa is trained entirely on human speech using a single NVIDIA GPU and can talk and listen simultaneously, similar to OpenAI's GPT-Live. Block claims emotion is built into the model rather than driven by text pipelines or voice tags, and shared examples showing the agent sounding tired, emotional, and natural. Dorsey said voice is the most natural way to interact with intelligence. Dessa is not publicly available; Block has invited safety researchers to test it, likely to examine risks such as voice agents deceiving people. Block sees on-device voice AI as a lower-cost way to power business and consumer experiences, building on Square's existing voice ordering.
Read sourceBlock Releases Dessa, a Local Full-Duplex Voice Model
Block, Jack Dorsey's research lab, has released Dessa, a voice model that runs locally on a single NVIDIA GPU and supports full-duplex conversation, allowing it to speak and listen simultaneously. The model is trained entirely on human speech and includes emotional expressions such as laughter, pauses, and crying. According to the company, its knowledge comes from conversational data rather than being paired with a large language model. Dessa can run on consumer-grade computers such as a MacBook M4 Max. It is not yet publicly available, and Block is recruiting security researchers to test it.
Read sourceShow 2 older updatesHide older updates
Jack Dorsey's Block launches Dessa, an AI voice model that runs on a MacBook
Jack Dorsey's Block has launched "Dessa," an AI voice model that Dorsey says was trained entirely on speech. According to the announcement, the model is small enough to run on a MacBook. The post presents the launch as a just-in development from Block, the company led by Dorsey. No further details were provided in the source item about the model's architecture, training data specifics, performance benchmarks, availability, or pricing. The claim about running on a MacBook is attributed to Dorsey and is the only technical detail given. The announcement positions Dessa as a compact, speech-trained voice model from Block.
Read sourceJack Dorsey's Block Releases Dessa, a Pure Voice-to-Voice AI Model
Jack Dorsey's company Block has released a voice AI model called Dessa, which the team says was trained entirely from scratch using only human conversations and is a true voice-to-voice model. Sample dialogues show natural interaction but occasional awkwardness and misidentification. The team says Dessa can run locally on a MacBook Pro M4 Max or higher configuration, and early access will be opened to a small group of AI safety researchers.
Read source