OpenAI launches GPT-Live-1 API with interruption handling, tool calls and phone agents
OpenAI has officially launched GPT-Live-1, a new API model designed to empower developers building full-duplex, real-time voice applications. This release signals a major shift in AI voice interaction, moving beyond traditional segmented processing toward fluid, natural conversation.
Technically, GPT-Live-1 unifies speech understanding and generation within a single model. By eliminating the latency caused by chaining separate speech-to-text, reasoning, and text-to-speech stages, it achieves near-instant response times. The model natively supports full-duplex interaction, enabling simultaneous listening and speaking while accurately managing interruptions, pauses, and background noise.

Beyond seamless interaction, GPT-Live-1 offers extensive customization. Developers can fine-tune tone, speed, and style via system prompts, while integrating backend models and intelligent agent frameworks. It features native speech recognition with alphanumeric understanding, keyword bias, and turn detection. Compatible with tools like Codex and OpenAI Presence, it powers enterprise queries, approved operations, and seamless handoffs to human agents, making it ideal for phone-based services like restaurant reservations and customer support.
Real-world results are impressive. Early adopters, such as language platform Speak, reported an 80% drop in mis-interruptions during learner pauses compared to round-based systems. Medical service platforms have also simplified their architectures and reduced code complexity. According to OpenAI, GPT-Live-1 surpasses GPT-Realtime-2.1 by 30 percentage points in Full Duplex Bench tests and leads Tau3 end-to-end voice agent benchmarks when paired with GPT-6Astra at moderate reasoning intensity.
Related article
Nvidia CEO Jensen Huang Dismisses AI Apocalypse Claims, Says 2030 Doomsday Probability Is Zero
In a recent CBS News interview, NVIDIA co-founder and CEO Jensen Huang dismissed warnings from certain researchers that artificial intelligence could lead to human extinction within a few years, labeling such apocalyptic narratives as exaggerated.“20
Moonshot: Can China’s 2.8 Trillion Parameter AI Rival US Tech?
In February 2026, Anthropic accused Chinese AI labs DeepSeek, Moonshot and MiniMax of “industrial-scale campaigns” to “illicitly extract Claude’s capabilities to improve their own models”. Credit: Alvan Nee/UnsplashChinese AI firm Moonshot has releas
Cozy 3.0 Launches With Multi-Person and Multi-Agent Collaboration Support
ByteDance has officially released Coze 3.0, introducing a novel approach to Agent creation while delivering major advancements in team collaboration and ecosystem integration.This update prioritizes collaborative efficiency, fully supporting flexible
Related Special Topic Recommendations
Comments (0)
0/500
OpenAI has officially launched GPT-Live-1, a new API model designed to empower developers building full-duplex, real-time voice applications. This release signals a major shift in AI voice interaction, moving beyond traditional segmented processing toward fluid, natural conversation.
Technically, GPT-Live-1 unifies speech understanding and generation within a single model. By eliminating the latency caused by chaining separate speech-to-text, reasoning, and text-to-speech stages, it achieves near-instant response times. The model natively supports full-duplex interaction, enabling simultaneous listening and speaking while accurately managing interruptions, pauses, and background noise.

Beyond seamless interaction, GPT-Live-1 offers extensive customization. Developers can fine-tune tone, speed, and style via system prompts, while integrating backend models and intelligent agent frameworks. It features native speech recognition with alphanumeric understanding, keyword bias, and turn detection. Compatible with tools like Codex and OpenAI Presence, it powers enterprise queries, approved operations, and seamless handoffs to human agents, making it ideal for phone-based services like restaurant reservations and customer support.
Real-world results are impressive. Early adopters, such as language platform Speak, reported an 80% drop in mis-interruptions during learner pauses compared to round-based systems. Medical service platforms have also simplified their architectures and reduced code complexity. According to OpenAI, GPT-Live-1 surpasses GPT-Realtime-2.1 by 30 percentage points in Full Duplex Bench tests and leads Tau3 end-to-end voice agent benchmarks when paired with GPT-6Astra at moderate reasoning intensity.
Nvidia CEO Jensen Huang Dismisses AI Apocalypse Claims, Says 2030 Doomsday Probability Is Zero
In a recent CBS News interview, NVIDIA co-founder and CEO Jensen Huang dismissed warnings from certain researchers that artificial intelligence could lead to human extinction within a few years, labeling such apocalyptic narratives as exaggerated.“20
Moonshot: Can China’s 2.8 Trillion Parameter AI Rival US Tech?
In February 2026, Anthropic accused Chinese AI labs DeepSeek, Moonshot and MiniMax of “industrial-scale campaigns” to “illicitly extract Claude’s capabilities to improve their own models”. Credit: Alvan Nee/UnsplashChinese AI firm Moonshot has releas
Cozy 3.0 Launches With Multi-Person and Multi-Agent Collaboration Support
ByteDance has officially released Coze 3.0, introducing a novel approach to Agent creation while delivering major advancements in team collaboration and ecosystem integration.This update prioritizes collaborative efficiency, fully supporting flexible





Home






