OpenAI launches GPT-Live-1 API at $0.05/minute
OpenAI has made GPT-Live-1, its full-duplex voice model already shipping in ChatGPT, available to developers at $0.05 per minute, replacing the stitched speech-to-text, LLM and text-to-speech pipeline with a single model.
GPT-Live-1 listens and speaks simultaneously, reasoning over incoming and outgoing audio in one model rather than chaining STT, an LLM and TTS — the handoffs OpenAI blames for latency and lost conversational timing. Deeper reasoning and tool calls are delegated to a backend text model such as GPT-6 Astra, Luna, or a third-party model, which developers choose and pay for separately; the $0.05-per-minute price covers only the front-end voice layer.
OpenAI reports GPT-Live-1 improves Full Duplex Bench performance by 30 percentage points over GPT-Realtime-2.1, with gains in turn-taking latency and interactive behavior. Paired with GPT-6 Astra at medium reasoning effort, it ranks first on Tau3, a benchmark for end-to-end voice-agent tasks. Language-learning app Speak found in early evaluations that the model cut interruptions by almost 80% versus previous turn-based systems, giving learners more time to think.
The API release adds system-prompt control over tone, pace and conversational style, native ASR transcripts and response text, keyword biasing, alphanumeric understanding, and telephony support for phone-based agents. Though not turn-based, the model natively supports turn detection for developers building around explicit turn boundaries. OpenAI says it is expanding beyond its previous small set of realtime voices to more accents, dialects and languages, with further additions "over the coming months."
The post names no specific languages or voices, gives no latency figures, and does not detail benchmark methodology. Enterprises can also access the model through OpenAI Presence via an account director.