Block's Dessa claims the most authentic full-duplex speech model ever built

Dessa: the most authentic full-duplex speech model ever built

Block has introduced Dessa, a conversation model trained exclusively on speech rather than text, which it bills as the most authentic full-duplex speech model ever built. The claim centers on full-duplex operation—listening and speaking at once—learned from audio alone. The available page, however, is a Block privacy notice covering data collection, sharing, cookies, and user rights, so the model's capabilities remain to be verified.