OpenAI launches GPT-Live-1 API for full-duplex voice, letting apps talk and listen simultaneously

OpenAI released GPT-Live-1 as a developer API: a single-model full-duplex voice system that beats GPT-Realtime-2.1 on interactivity, latency, and tool calling at $0.05 per minute, and is already in production at Speak and Yelp.

OpenAI launched GPT-Live-1 in the API, a full-duplex voice model that listens and speaks simultaneously in a single model, replacing chained STT-LLM-TTS architectures that introduced brittle handoffs and latency. It delegates reasoning and tool calls to backend models such as GPT-6 Astra or third-party options, and it already powers ChatGPT voice. Benchmarks reported by The Decoder show 80.1% on full-duplex interactivity versus 45.4% for GPT-Realtime-2.1 (OpenAI's own announcement cites a roughly 30-point improvement on Full Duplex Bench), turn-taking latency of 0.8 seconds versus 1.4 seconds, tool-calling accuracy of 87% versus 60%, and a 32% pass rate on a banking voice-support benchmark versus 12.4%. Paired with GPT-6 Astra at medium reasoning effort, GPT-Live-1 ranks #1 on the Tau3 benchmark. The API costs $0.05 per minute for the front-end voice layer, ships twelve new voices, and provides native ASR transcripts, response text, keyword biasing, telephony support, and expanded voice and language options. Early adopters: Speak reported nearly 80% fewer interruptions in language tutoring, and Yelp uses the model for phone-based reservations with reported better call handling.

• GPT-Live-1 is a single-model full-duplex voice system exposed via API; it already powers ChatGPT voice.
• Full-duplex interactivity score: 80.1% versus 45.4% for GPT-Realtime-2.1 (OpenAI cites roughly +30 points on Full Duplex Bench; sources' framing differs slightly).
• Turn-taking latency of 0.8 seconds versus 1.4 seconds for GPT-Realtime-2.1.
• Tool-calling accuracy of 87% versus 60%; banking voice-support benchmark pass rate of 32% versus 12.4%.
• Ranks #1 on Tau3 when paired with GPT-6 Astra at medium reasoning effort; reasoning and tool calls are delegated to backend models, including third-party options.
• Pricing: $0.05 per minute for the front-end voice layer.
• Ships twelve new voices with native ASR transcripts, response text, keyword biasing, telephony support, and expanded voice and language options.
• Production deployments: Speak reports nearly 80% fewer interruptions in language tutoring; Yelp uses GPT-Live-1 for phone-based reservations.