More natural voice conversations with GPT-Live 1, 12 new voices, and GPT-6 Sol as the analysis default

• Perspective now supports GPT Live 1 through our opt-in Live (Beta) voice conversations. With full duplex audio, the agent can listen while it speaks. If a participant remembers an important detail, corrects themselves, or changes direction while the agent is replying, they can say it right away. The agent can stop, listen, and follow the new context; it can also give a quiet acknowledgment while the participant continues. That gives people more room to finish a thought, and their answers stay paired with audio so you can listen back. Workspace owners and admins can try Live (Beta) under Settings → Model → Voice interviewer. The existing Realtime voice engine remains the default unless you switch.
• Give each agent a voice that fits its audience. Live (Beta) adds 12 voices to the agent voice picker, including Australian, Irish, Filipino, Southern U.S., and Brazilian Portuguese options. You can hear a sample before choosing how your agent will sound in a conversation.

GPT-6 Sol is now the default model for analysis sessions and scheduled insights. It applies when you start a session or reopen one; you can still choose another model for analysis.