Adventists AI releases its Sophon-Realtime v1 voice models under Apache 2.0
Sophon-Realtime v1 is a 32-billion-parameter full-duplex speech model family. It includes an expressive Tag variant, a low-latency Flash variant, and an end-to-end research preview. The team reports a 0.31-second delay to first audio for Flash and says both Tag and Flash run on a single 80 GB GPU.
Visit original source