Voice platform engineer tests GPT-Live-1 vs Gemini 3.8 Live on real phone calls
VladimirSamukov · reddit · 2026-10-01
An engineer at voice AI orchestration platform Voximplant shares first impressions of GPT-Live-1 and Gemini 3.8 Live for phone voice agents.
GPT-Live-1
- Feels like a real conversation: two-way interruptions, finishes its sentence while you talk without losing context, even interrupts the user once to flag a duplicate task.
- Catches mid-conversation additions naturally and offers timely backchannel ("uh-huh") — crucial since silence makes callers hang up.
- In their tests, emotional TTS added little over standard TTS; conversation management is where it stands out.
Gemini 3.8 Live
- Function calls run in the background, so conversations don't stall during API lookups.
- Extended Thinking narrates its progress aloud in real time.
- Seamless mid-conversation language switching, a big deal for India and LatAm use cases.
The author also notes that at high outbound call volumes, cost calculations shift significantly. Both models are supported on Voximplant.
More from coding & agent
- A 5-second talking-head loop in 6GB VRAM: local graph, hosted lip-sync API — Extreme-Shock1930 · 2026-10-01
- My agents burned video credits on bad frames, so now they must ask before spending — coolxeo · 2026-10-01
- ComfyUI plugin Continuity update: in-shot screen replacement, image-to-3D, Qwen Image 2.1 — Fine_Rhubarb3786 · 2026-10-01
- Matt Pocock on 4-6 hour AI coding tickets: check feedback loops, then run /retro — mattpocockuk · 2026-10-01
- Redditor hunts for open-source Claude Cowork alternative, finds sparse field — caenum · 2026-10-01
- Letta Code kills blocking AskUserQuestion: async agents still need a stop default — sujingshen · 2026-10-01