By No Code MBA · Facts checked
- Provider
- Block
- Announcement date
- Status at last check
- Limited access or preview
- Model family
- Dessa
- Capabilities
- Audio
What changed
Dessa is a full-duplex speech-to-speech model from Block's AI research lab that listens and talks at the same time. Block says it was trained only on human speech sounds, with no text and no pretrained models, on a single NVIDIA GB200 NVL72. Block reports that it learned language from hundreds of times fewer words than LLMs, runs faster than real time, handles laughter, sarcasm, pauses and backchanneling, and can run locally on a MacBook Pro M4 Max or better.
Access and availability
There is no public API or download. A small group of AI safety researchers has early access, and other researchers can join a waitlist or request access with an institutional email via the Dessa page.
Official access documentationWhat you can build
If you can access the preview, evaluate interruption handling, latency, and emotional nuance in a short spoken task, with human review before relying on responses.
Limitations
Block publishes no parameter count, latency figures, benchmarks or license terms, only demo recordings. The company describes learning from sound alone as new territory and asks outside researchers to find risks it has not anticipated.