Block, led by Jack Dorsey, has unveiled Dessa, a new voice AI model that stands out with its remarkably human-like expressions, including laughter, pauses, and emotional inflections such as cries. Unlike many speech models that depend heavily on text-based inputs to generate speech, Dessa is trained purely on human speech data using a single NVIDIA GPU rack. This full-duplex model can simultaneously listen and speak, mirroring natural conversation flow, comparable to OpenAI’s GPT-Live voice model introduced earlier in 2026.

Dessa differentiates itself by embedding emotion intrinsically into the speech generation process, rather than relying on external voice tags commonly used to simulate emotion in other models like ElevenLabs v4. Block’s research team demonstrated this capability with dialogues where the AI expresses feelings such as tiredness or becoming emotional, enhancing the authenticity and nuance of interactions. The company emphasizes the importance of voice as a natural medium for intelligence interaction, aiming to avoid the loss of emotional nuance that occurs when voice is converted to text in other systems.

The model is designed to run locally on consumer-grade hardware, such as a MacBook powered by the M4 Max chip or better, which could reduce costs and enable more widespread business and consumer applications. While there is currently no public release of Dessa, Block has invited safety researchers to evaluate the model’s potential risks, especially regarding scenarios where voice AI could mislead users. This move suggests Block’s cautious approach toward responsible deployment amid rising concerns about the trustworthiness of AI voice agents.

Block, traditionally focused on financial services and small business solutions, could leverage Dessa to expand voice AI capabilities, complementing existing tools like Square’s AI-powered voice ordering system. The new model could open doors for richer voice-based interactions in commerce and beyond, signaling a significant step forward in making speech AI that feels genuinely human. For now, Dessa remains in a testing phase, with interest growing over how it might reshape user engagement with voice technology.