GPT-Live goes full-duplex. Voice agents stop waiting their turn.
OpenAI's GPT-Live listens and speaks at once and hands hard questions to GPT-5.5 mid-sentence. Here's what full-duplex means for the voice agent on your phone line.
OpenAI shipped GPT-Live on July 8 — a pair of voice models, GPT-Live-1 and GPT-Live-1 mini, built to talk over the phone the way a person does. The headline feature is full-duplex: the model listens and speaks at the same time, so it can be interrupted, backchannel with an "mhmm," and shut up while you think. If you've ever built or bought a voice agent and watched it steamroll a caller because it couldn't tell they'd started talking, this is the fix. And it changes the math on what a voice agent can do on your business line.
What actually happened
The old voice stack was three parts bolted together: speech-to-text, a language model, then text-to-speech. It had to wait for you to finish talking before it could even start thinking. TechCrunch reports GPT-Live collapses that into a single full-duplex model that "can speak and listen at the same time," enabling natural interruptions and live translation. OpenAI says more than 150 million people already talk to ChatGPT through voice and dictation.
Here's the clever part. GPT-Live stays fast by not trying to be smart on its own. When a question needs real reasoning, search, or an agentic step, it hands off to GPT-5.5 in the background and keeps the conversation going while the answer comes back. GPT-Live-1 mini replaces Advanced Voice Mode for free users; GPT-Live-1 is the default on paid plans. It's rolling out across iOS, Android, and the web, and OpenAI says a developer API is planned — though pricing and API timing aren't public yet.
Why it matters for your business
The failure mode of every cheap phone bot is the awkward pause and the interruption it can't handle. Full-duplex kills both. That moves voice agents from "obviously a robot" to "good enough for your actual customers" — the after-hours line, the order-status call, the reschedule. We've built exactly this on LiveKit and Twilio for a retail client's three stores, and turn-taking was the hardest part. Better base models make that easier, not automatic.
Two things stay true. First, the model is a component, not the product — GPT-Live delegating hard questions to GPT-5.5 is a preview of your build, where the voice layer answers fast and routes anything real to a system that knows your bookings, inventory, and FAQs. Second, don't marry your phone line to one vendor's app. Build the agent so the voice model is a swappable part. When the next full-duplex model ships cheaper — and it will — you change one layer, not your whole system.
Key takeaways
- GPT-Live-1 and GPT-Live-1 mini are full-duplex — they listen and speak at once, so they can be interrupted and pause naturally
- It replaces the old speech-to-text → LLM → text-to-speech pipeline that had to wait for you to stop talking
- GPT-Live delegates hard questions to GPT-5.5 mid-conversation — fast voice layer, deep reasoning behind it
- A developer API is planned but not yet priced; the winning build keeps the voice model swappable
Want a phone line that actually holds a conversation? We build voice agents around how your business answers — booking rules, order status, FAQs, clean handoff to a human — with the model layer kept swappable so you're never locked to one vendor. See what we build or tell us what your line handles.
Sources: TechCrunch — OpenAI releases new voice models for more natural live conversations, SiliconANGLE — OpenAI launches GPT-Live voice model series.
- #voice-ai
- #gpt-live
- #ai-agents
- #automation
- #openai
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
ServiceTrade bought an agentic billing agent. Own your last mile.
ServiceTrade acquired Mura to automate field-service order-to-cash with agentic AI. Agentic billing has reached the boring, high-value core of service ops.
Read itPrime Intellect raised $130M so you can train your own agents
Prime Intellect hit a $1B valuation selling infrastructure to train AI agents on your own data instead of renting a frontier lab. Own the optimization loop.
Read it