GPT-Live
AI overview
Sign in and the AI will write an overview from our coverage.
Headlines · 8
- How Voice Agents Are Evolving Toward Full-Duplex Interaction
A supplied report describes three voice-agent architectures: cascading pipelines, end-to-end omni-modal models, and full-duplex systems. It cites Moshi, GPT-Live, and Step-Audio R1 as representative approaches. The report identifies a central design challenge: supporting timely, interruptible interaction while allowing deeper reasoning, and presents these as architectural trade-offs rather than a linear progression. It also argues that voice enables continuous, hands-free agent workflows rather than merely text-to-speech.
掘金 · 人工智能 · 🔥 0 - OpenAI to retire two ChatGPT voice modes on December 11
OpenAI will retire ChatGPT’s Advanced and Standard voice modes as selectable options on December 11, 2026, according to TechRepublic, while retaining conversations created with them. Accounts using either mode will switch automatically to GPT-Live, which OpenAI says can listen and speak simultaneously, support interruptions and pass complex requests to other AI models during an active conversation.
TechRepublic · 🔥 0 - OpenAI to Retire Two ChatGPT Voice Modes
OpenAI will retire ChatGPT’s Standard and Advanced Voice modes on December 11, leaving Live as the only voice mode across all subscriptions, according to the supplied reports. OpenAI says nearly all ChatGPT users are already using Live, while a smaller but vocal group objects to losing the older modes. Some users prefer the way Standard and Advanced respond or have become attached to particular voices, and Live does not provide the same experience even when a similar voice is selected. Reddit users are sharing a petition and urging people to send feedback directly to OpenAI before the change.
TechRadar · 🔥 0 - Shizuku-Duplex Adds Full-Duplex Japanese Conversation
Japan-based Shizuku AI announced Shizuku-Duplex, a full-duplex model designed for Japanese conversation that can begin processing before a request is complete. The model combines a foreground interaction model with a background model for asynchronous complex requests, supports audio, text-monologue and action outputs, and uses 80-ms interaction frames with context compression every five minutes. Shizuku AI says it achieved the highest conversational naturality among six tested Japanese models and ranked second to OpenAI’s GPT-Live in response quality, but no test method, scores or independent verification were provided.
IT之家 AI · 🔥 0 - Why Ten Minutes of Voice Chat Cannot Be Priced by Transcription Minutes
An analysis of OpenAI's official documentation explains that conversational Realtime usage is metered mainly by the input and output tokens of each Response, while the GPT-Live voice component is billed by active-session seconds, with background models and tools billed separately. It presents two independently constructed calculations run with a local Python Decimal script: one synthetic gpt-realtime-2.1 Response priced at $0.2092 under prices checked on October 10, 2026, and a separate GPT-Live voice calculation for 601 seconds at a $0.05 per-minute rate, yielding about $0.50083333. The author states that no device was connected, no model was called, no billing was incurred, and no real usage data or actual invoice was obtained; the script only verifies the arithmetic of the constructed inputs, and missing components such as transcription and backend are reported as null rather than zero.
掘金 · 人工智能 · 🔥 0 - OpenAI launches GPT-Live outbound telephony API
OpenAI promoted the GPT-Live outbound telephony API to general availability. Developers can use a Live voice agent to place phone calls through any SIP trunk provider, including Telnyx or Twilio, with support for HD Voice and sending or receiving DTMF keypad events.
Greg Brockman · OpenAI President · 🔥 0 - OpenAI Launches Decisions API for Fast Structured Decisions
OpenAI has made the Decisions API generally available, powered by GPT-6 Luna for classification, routing, scoring, and other bounded decisions across text, images, and structured inputs. OpenAI says it can respond up to 10 times faster than GPT-6 Luna through the Responses API; pricing is $0.10 per million input tokens, with cached inputs at $0.01 per million and no output-token charges. The endpoint adds native JSON input, image instructions, up to 400 questions per request, prompt caching, US and EU inference residency, and native GPT-Live integration.
OpenAI · 🔥 0 - Azeem Azhar spent a week working by voice alone with OpenAI's GPT-Live
Azeem Azhar, founder of Exponential View, left his laptop at home in Amsterdam while travelling to London and spent the following week doing his daily work entirely by voice through ChatGPT on his phone. According to AIbase, he used OpenAI's GPT-Live feature to direct a backend AI agent by speaking and listening, comparing the experience to Tony Stark's Jarvis. He said the setup handled most of his core tasks, which he presents as evidence that voice-driven agents can carry real work.
AIbase AI新闻 · 🔥 0
Experience and discussion from the community
Share my GPT-Live experienceAsk about GPT-Live
Nobody has shared their experience with GPT-Live yet.