OpenAI releases GPT-Live-1.1 with multilingual voices

OpenAI has launched GPT-Live-1.1 along with a smaller version named GPT-Live-1.1 Quip, updating its real-time conversational model across ChatGPT, Codex, and dots, with API integration planned shortly. The upgrade introduces localized voice options for French, German, Hindi, Japanese, Korean, and Mexican Spanish, alongside a new English voice named Bubbie.

Beyond voice options, the release fixes a common pain point in full-duplex conversational systems. Previously, the original GPT-Live-1 repeatedly uttered phrases like "checking" while invoking external tools like code execution or web searches. GPT-Live-1.1 expands its library of spoken transition cues to reduce monotonous waiting audio. Although tool execution speeds remain unchanged, the update enhances the audible layer of the model's decision loop, which enables simultaneous listening, speaking, mid-sentence turn-taking, and backchannel responses like "mhmm".

Unlike older voice systems that relied on a three-part pipeline converting speech to text, running a language model, and rendering text-to-speech, the Live architecture processes continuous audio natively. Pricing and technical specifications for GPT-Live-1.1 have not yet been disclosed, though original GPT-Live-1 API pricing was set at $0.05 per minute across 12 voice choices. Custom voices remain gated behind enterprise sales, and OpenAI retains built-in guardrails against voice impersonation.

For developers building multilingual applications, voice-driven Codex workflows, or automated support agents, the update promises less mechanical interactions without requiring structural prompt alterations. Once the API opens, teams migrating from earlier builds should benchmark tool latency separately from spoken status cues, verify localized voice availability across target regions and account tiers, and review updated pricing schedules and model identifiers before shifting production traffic.