Research preview
Full-duplex, speech-to-speech research model.
Checked 2026-10-24 · Source ↗
Speech model field notes
A clearer view of
what comes next.
Dessa is a speech model developed by Block, Inc. Read the source ↗.
Understand what you can try today, what your Mac needs, and what has yet to be published.
Release tracker · 2026-10-24
Promising research. Clear boundaries.
Public waitlist
The button did not respond in Chrome/macOS on 2026-10-24. Researcher applications are separate.
Checked 2026-10-24 · Source ↗
App
No public app package verified on the checked page.
Checked 2026-10-24 · Source ↗
API
No public API documentation verified.
Checked 2026-10-24 · Source ↗
Weights & license
No public weights package or distribution license verified.
Checked 2026-10-24 · Source ↗
Listen at the sourceConversation that
Conversation that
keeps listening.
Block describes a model trained on human speech that can listen and speak at the same time. Its recordings illustrate timing, interruptions and expression. Those demonstrations are publisher examples, not independent benchmarks.
Visit Block’s release page ↗Before you plan a build
No published model size, minimum memory, API price or commercial license is verified here. Avoid treating a demonstration as a shipping product.
Access guide →Keep a place in your inboxKnow when the
Know when the
status changes.
Request updates about APIs or weights. These emails come from this site’s operator, Jianming Duan. The optional budget question concerns future tools by Jianming Duan, not an offer of Dessa access or hosting. They do not enroll you in Block’s waitlist.
Explore the Mac launcher prototype →