Block introduces Dessa, a full-duplex speech model trained only on conversation
Block says Dessa learns from human conversation without text or pretrained models, can listen while speaking, and runs locally on consumer hardware.
AT A GLANCE
Key points
- Trained from scratch on human conversation, according to Block.
- Full-duplex design listens and speaks at the same time.
- Runs locally on an M4 Max MacBook Pro or better.
What is changing?
Block has introduced Dessa, a conversational model it says was trained from scratch exclusively on human speech. According to the company, its knowledge does not come from text or a pretrained language model; it maps audio input directly to audio output.
Dessa operates in full duplex: it can keep listening while the user speaks and respond at the same time, including interruptions and short backchannels as part of the conversational flow. Live recordings on Block’s official page show behaviors such as recalling a movie title, laughing during a joke and changing emotional tone.
Block says inference runs faster than real time and the model can operate locally on a MacBook Pro with an M4 Max chip or better. General access is not available yet; the company plans early access for a small group of AI safety researchers.
Explore the original sources
Source date: 23 October 2026.
Published by AI Radar: 23 Oct 2026, 23:11 (Istanbul, UTC+3). Updated: 23 Oct 2026, 23:12.
FOLLOW THE TOPIC
Topics and tags
Share your discovery.
Send this story to a friend.
Reader comments
What do you think about this development? Share your experience or question. Comments are published after editorial approval.
Your name/nickname and comment become public if approved. Read: Privacy notice.
Loading comments…