# Block releases Dessa, a full-duplex speech model trained only on speech

> Original title: Block Unveils Dessa, a Full-Duplex Speech Model Trained Only on Speech

- Source: MarkTechPost
- Published: 2026-10-29T07:50:39.000Z
- Bad Influence: https://badinfluence.news/items/zhu375x0xpu83k3n5bwptseli
- Original: https://www.marktechpost.com/2026/10/29/block-dessa-full-duplex-speech-model/

## Summary

Block's AI research lab announces Dessa, a speech-to-speech model that listens and talks at the same time, trained only on human conversation audio with no text or pretrained models. Block says it trained Dessa on one NVIDIA GB200 NVL72 rack, compared with 1,016 H100 GPUs used for Kyutai's Moshi, and that it runs locally on a MacBook Pro M4 Max faster than real time. The research preview has no published benchmarks, parameter count, weights, API, or price, and access is by waitlist.