One chart mapping voice AI: S2S vs cascaded, half- vs full-duplex
rdesh26 · x · 2026-11-17
A BlueJay Labs infographic shared by @farazmsiddiqi maps the voice AI landscape along two axes: cascaded (ASR→LLM→TTS) vs speech-to-speech, and half- vs full-duplex. Key takeaways:
- Full-duplex (listening while talking) is currently only achievable with hosted S2S models like GPT-Live-1, Seeduplex, MOSS-Live 1.0, and Phonic-1; no hosted cascaded system supports it yet.
- Open-weight S2S full-duplex options include Moshi, PersonaPlex 7B, MiniCPM-o 4.5, and Haru-4B.
- The half-duplex S2S quadrant lists Grok Voice, Gemini Live, Realtime 2.1, Nova Sonic, Qwen Omni, plus open weights like Qwen3- Omni 30B and Kimi-Audio 7B.
- Platform options (Vapi, Retell, LiveKit, Twilio, Amazon Connect, etc.) are mapped per architecture, making it a handy selection cheat sheet.
More from coding & agent
- Boring Marketer open-sources 32 marketing agent skills and launches forward-deployed service — boringmarketer · 2026-11-17
- Pybot experiment write-up coming: how the autonomous GitHub bot actually ran — chrisalbon · 2026-11-17
- Running a company in 2031: a meme of one human in a meeting with nine Claude agents — andrew_n_carr · 2026-11-17
- Perplexity to upgrade its Slack agent Computer into a persistent assignable Teammate — AravSrinivas · 2026-11-17
- Subagents considered harmful: lossy summaries cost more and drop context — rseroter · 2026-11-17
- Boring Marketer open-sources 32 Claude skills covering every AI marketing use case — boringmarketer · 2026-11-17