GPT-Live-1.1 Launches with Full-Duplex Voice AI

OpenAI has officially released GPT-Live-1.1, a full-duplex voice model designed for natural, real-time multilingual conversations. The model processes speech input and output simultaneously, enabling interruptions and overlapping dialogue that mirror human conversation patterns. Available through OpenAI's Realtime API, the release targets developers building voice-first applications requiring low-latency interaction.
Full-Duplex Architecture
GPT-Live-1.1 operates in full-duplex mode, meaning it can listen and speak at the same time without requiring turn-based exchanges. This architecture allows users to interrupt the model mid-response, with the system dynamically adjusting its output based on new input. The model maintains context across interruptions, preserving conversation flow even when multiple speakers overlap.
Developers can configure latency thresholds and voice activity detection parameters through the API. The system supports streaming audio input and output, with configurable buffer sizes to balance responsiveness against processing overhead. Early implementations report sub-200ms response initiation times in optimal network conditions.
Multilingual Support
The model handles multiple languages within a single conversation session, automatically detecting language switches and responding in the appropriate language. Supported languages include English, Spanish, French, German, Italian, Portuguese, Dutch, Polish, Russian, Japanese, Korean, and Mandarin Chinese.
Unlike earlier voice models that required language pre-specification, GPT-Live-1.1 adapts dynamically. Users can switch languages mid-conversation without re-initializing the session. The model preserves semantic context across language boundaries, maintaining topic continuity when users alternate between languages.
Release Date and Availability
Officially launched on November 19, 2026, GPT-Live-1.1 is immediately available to all OpenAI API customers with Realtime API access. The model appears in the standard model endpoint list alongside text and vision models. Developers can test the system through OpenAI's Playground interface before production deployment.
Pricing follows OpenAI's existing Realtime API structure, charging per audio minute processed. The company has not disclosed whether GPT-Live-1.1 carries different rates compared to the original GPT-Live model. API documentation includes code samples in Python, Node.js, and cURL for quick integration.
Use Cases and Integration
Target applications include voice assistants, customer support bots, language learning platforms, and accessibility tools. The full-duplex capability addresses a key limitation in prior voice AI systems, where users had to wait for the model to finish speaking before providing input.
Several developers on Hacker News noted the model's improved interrupt handling compared to the original GPT-Live release. One commenter highlighted reduced latency jitter in multi-turn conversations. Another pointed to the multilingual feature as critical for global deployment without maintaining separate model instances per region.
What This Means
GPT-Live-1.1 represents a step toward more natural voice interaction in AI systems. By supporting simultaneous two-way communication and seamless language switching, the model removes technical barriers that previously made voice AI feel robotic or constrained. For developers, the release offers a production-ready tool for building conversational experiences that more closely approximate human dialogue patterns. The immediate availability through existing API infrastructure lowers adoption friction, positioning full-duplex voice as a standard capability rather than an experimental feature.
on Emergent today





