GPT-Live-1 Real-World Test: Most Natural Voice, Shaky Instruction-Following

AI phone agent firm ThunderPhone tested OpenAI's full-duplex voice model GPT-Live-1 on real phone lines with an insurance script. The voice was the most natural among peers, but instruction-following failed repeatedly—even on a simple "yes".

2026-09-13 ~ 2026-09-14 · 2 related posts

• GPT-Live-1 first impressions: most natural voice model yet, but instruction following is unreliable — kolchinski · 2026-09-13
• GPT-Live-1 tested on real phone calls: natural speech but serious instruction-following flaws — kolchinski · 2026-09-14