OpenAI rolls out GPT-Live-1.1 to all ChatGPT Voice users with 13 new voices

OpenAI's latest voice model, GPT-Live-1.1, is now rolling out globally. The update adds 13 new natural voices across English, French, German, Hindi, Japanese, Korean and Mexican Spanish, with more natural regional accents. It also cuts conversational latency and introduces clearer audio cues when the assistant runs a live web search, a common source of awkward pauses in earlier versions.

The rollout covers ChatGPT Voice, Codex and dots. Free-tier users get a streamlined variant called Quip. OpenAI says the model incorporates protections from its October Preparedness Framework against multi-turn jailbreaks, a known weak spot for voice assistants because spoken conversations are harder to filter. API access is listed as coming soon, which matters most to developers building production voice agents.

The competition is direct. Two days earlier, xAI announced Grok Voice 3.0, a native speech-to-speech model that scored 86.9 on Artificial Analysis's Speech-to-Speech Index and 97% on Full Duplex Bench for handling interruptions. xAI says it already handles customer routing for Tesla and SpaceX. Google's Gemini Live and Apple's new Siri-powered HomeView also compete for voice share, so latency and multilingual naturalness have become the battleground.

What to watch: OpenAI has not published benchmark numbers for GPT-Live-1.1, so head-to-head comparisons with Grok Voice 3.0 remain anecdotal. API pricing and availability will decide whether enterprises building call-center and agent products adopt it, and whether the multi-turn jailbreak defenses hold up under red-teaming.