GPT-Live-1 @openaidevs is now supported in LiveKit Agents. It's full-duplex, so it listens and speaks at the same time instead of taking turns, and it hands reasoning and tool calls off to a backend model so a slow lookup doesn't leave your agent sitting in dead air. Same `AgentSession` API. New plugin: `GPTLiveModel`. Get started:

GPT-Live 1 is on AI Gateway. • Full duplex audio model: listens while speaking, so you can interrupt. • Client delegation: run any text model in the background. 𝚘𝚙𝚎𝚗𝚊𝚒/𝚐𝚙𝚝-𝚕𝚒𝚟𝚎-𝟷

GPT-Live-1 (Astra, medium) costs $5.83 and GPT-Live-1 (Sol, low) costs $4.47 per hour of input audio on our fixed 40-question Big Bench Audio pricing subset, compared with $4.80 for Grok Voice Think Fast 2.0 High and $10.75 for GPT-Realtime-2.1 High. These figures include GPT-Live-1’s voice session charge and delegated backend text model token usage at standard rates.

GPT-Live-1 (Astra, medium) averages 1.34 seconds to first audio on Big Bench Audio, compared with 1.24 seconds for GPT-Live-1 (Sol, low), 0.70 seconds for Grok Voice Think Fast 2.0 High and 1.21 seconds for GPT-Realtime-2.1 High.

GPT-Live-1 (Sol, low) ranks #3# in conversational preference at 1,053 Elo with 90.9% Task Success Rate, while GPT-Live-1 (Astra, medium) ranks #4# at 1,048 Elo with 87.4% task success. Gemini 3.1 Flash Live Minimal leads preference at 1,096 Elo but achieves 74.6% task success, while Grok Voice Think Fast 2.0 High leads task success at 94.6% with 1,011 Elo. GPT-Live-1 (Sol, low) has higher point estimates on both measures than GPT-Live-1 (Astra, medium), although their confidence intervals overlap.

GPT-Live-1 (Astra, medium) leads Tau Voice at 67.9%, followed by GPT-Live-1 (Sol, low) at 59.3% and Grok Voice Think Fast 2.0 High at 56.5%, driving its overall Index lead. Both GPT-Live-1 configurations are currently based on one trial. On Big Bench Audio, they achieve 90.1% and 89.0%, respectively, behind Grok at 97.2%. On our Full Duplex Bench subset, they score 94.9% and 97.3%, respectively; this benchmark is reported separately from the Index.

GPT‑Live is incredibly well-executed. The short cycle times make the model smarter because you are injecting your own intelligence into the process more deeply. This model has its own intelligence toggle, I set it to High.