Open-source full-duplex speech model caramel-1 sounds human but lacks coherence

Developer naklecha released caramel-1, a 413M-parameter open-source full-duplex speech model trained purely on audio; it sounds remarkably human but cannot yet sustain coherent conversations.

2026-11-11 ~ 2026-11-12 · 2 related posts

• caramel: open-source speech-only conversational model sounds human but can't hold a coherent chat yet — realmrfakename · 2026-11-11
• Open-source full-duplex speech model caramel-1: 413M params, sounds human but can't hold a conversation — alexcovo_eth · 2026-11-12