The company says it was trained from scratch on human conversation audio only, without text or pretrained models. It describes Dessa as able to listen and speak simultaneously, handle interruptions and backchannel during a conversation.
According to Block, inference runs faster than real time and the model can run locally on a MacBook Pro M4 Max or better. The company also says training used one NVIDIA GB200 NVL72. The release page includes recorded conversations, including one with GPT-Live-1.
Dessa is not presented as a generally available API or open model. Block offers a waitlist and says a small group of AI safety researchers can seek early access. The source does not state model size, training-data volume or provenance, licensing, pricing, API availability, or independent evaluations of its conversational performance and novelty.