Key takeaways
- OpenAI released GPT-Live-1.1 in the API on 20 November 2026, a full-duplex voice model with improved multilingual accents and transcription.
- A new standby mode bills idle WebRTC time at $0.02 per minute instead of $0.05, a 60% cut for time when nobody is speaking.
- Active voice time still costs $0.05 per minute, the same rate as the earlier GPT-Live-1 model.
- Developers can call gpt-live-1.1 in voice applications through the v1/live/sessions endpoint, over WebRTC, WebSocket or SIP.
- Standby works on WebRTC sessions: after a configurable idle timeout of 5 to 60 seconds, inference pauses while the connection stays open.
What is GPT-Live-1.1?
GPT-Live-1.1 is OpenAI's newest full-duplex voice model, now available in the API. Full-duplex means the model can listen while it speaks, much like a person in a real conversation. It can be interrupted, it can hear a caller start talking while it is still answering, and it does not need the conversation to take turns in a rigid order.
The model is the same voice engine that has powered ChatGPT Voice since October 2026, now offered to developers building their own voice applications. OpenAI describes improved multilingual accent handling and clearer transcription of conversations in more than one language, which matters for customer support and other services used by callers around the world.
Developers reach it through the v1/live/sessions endpoint, over WebRTC, WebSocket or SIP. The model name to call is gpt-live-1.1. The update appeared in OpenAI's API changelog dated 19 November 2026, and the model became available in the API on 20 November.
How standby mode works
The most important part of the release is not the model itself but the way idle time is billed. Voice agents spend a large part of many calls doing nothing: waiting while a caller finds an order number, listening to hold music, or sitting silent between questions. Until now, every minute of a live session was billed at the full active rate, even when nobody was speaking.
Standby mode changes that for WebRTC sessions. Developers set two session settings. One, the wake policy, makes the session wait for speech before starting inference. The other, the idle timeout, sets how long the model stays active after the last speech, from 5 to 60 seconds. Once that timeout expires, OpenAI pauses inference while keeping the connection open. When someone speaks again, the session wakes and resumes.
While a session is on standby, the connection and the session ID are kept, and the conversation history is preserved. The microphone stays on, so a single word can wake the model. There is a small delay as inference restarts, but the call itself never drops.
What the price cut is worth
The numbers are simple. Active voice time costs $0.05 per minute, and standby time costs $0.02 per minute, billed per second. That is a 60% reduction for idle minutes. Ten idle minutes now cost $0.20 instead of $0.50. A 30-minute call with ten minutes of talking and twenty minutes of quiet costs about $0.90 instead of $1.50.
For a business running a support line, the savings add up quickly. An eight-hour support line that is idle for much of the day could save a meaningful amount each month, and smaller teams that could not afford a voice agent running all day may now find the numbers work. The standby rate applies to both GPT-Live-1 and GPT-Live-1.1 on WebRTC sessions, so existing users of the earlier model benefit too.
There are limits worth noting. Standby is a WebRTC feature, so phone and WebSocket calls still pay the full rate while idle. The voice session fee is also only part of the total bill: backend usage, such as the tools and systems the agent calls on, is billed separately and is unchanged. Anyone estimating costs should work from their own call logs, measuring how much of their traffic is actually idle, rather than assuming every minute will be billed at the standby rate.
Why voice agents are getting cheaper
Voice AI has been one of the fastest-growing parts of the AI market, but cost has been a constant problem. A text chatbot can sit idle for free, but a voice model in a live session was traditionally billed for every second it was connected. That made always-on voice assistants expensive, especially for small businesses, and it pushed developers to hang up quickly or to use cheaper, lower-quality speech systems.
Standby mode is a small technical change with a large commercial effect. It removes the penalty for leaving a voice agent connected during the quiet parts of a conversation, which is how real conversations actually work. People pause, think, look things up and get interrupted. A pricing model that charges full price for silence punished exactly that behaviour.
The release also shows how quickly voice models are becoming infrastructure rather than a novelty. OpenAI is competing with a growing field of voice providers, and pricing for idle time is a natural place to compete. For developers, the practical effect is that a voice agent that is mostly waiting is now far cheaper to keep running around the clock.
What developers should check
Before switching, developers should confirm a few things. First, standby only helps on WebRTC sessions, so teams using phone lines or WebSocket connections will not see the saving unless they change how they connect. Second, the idle timeout is configurable between 5 and 60 seconds, so teams can choose how quickly a session goes quiet. A short timeout saves more money but may add more wake-up delays, while a longer one is smoother but cheaper by less.
Third, the model's quality claims have not been independently tested yet. OpenAI says accents and multilingual transcription are improved, but there is no outside benchmark confirming how large the improvement is. Teams building for callers in many languages should test the model on their own calls before relying on it.
Finally, the active rate is unchanged at $0.05 per minute, so a voice agent that talks constantly costs exactly what it cost before. The saving only appears in the idle minutes. Teams whose agents are busy most of the time will see little change, while teams whose agents wait a lot will see the biggest benefit.
Conclusion
GPT-Live-1.1 is a useful, practical update rather than a dramatic new model. The voice engine is better at accents and multilingual transcription, and developers can now call gpt-live-1.1 in their own voice applications. The change that matters most is standby mode: idle WebRTC time now bills at $0.02 per minute instead of $0.05, a 60% cut that makes always-on voice agents far cheaper to run. For anyone building a support line, a booking assistant or a voice agent that spends its day waiting, the numbers now work much better.
Frequently asked questions
How much does GPT-Live-1.1 cost?
Active voice time costs $0.05 per minute, and standby time on WebRTC sessions costs $0.02 per minute, billed per second. The active rate is the same as the earlier GPT-Live-1 model. Backend usage for tools the agent calls is billed separately.
What is standby mode?
Standby mode pauses inference during quiet periods in a WebRTC voice session while keeping the connection open. After a configurable idle timeout of 5 to 60 seconds with no speech, the session goes quiet at the lower $0.02 per minute rate, and it wakes automatically when someone speaks.
Does standby work on phone calls?
No. Standby applies to WebRTC sessions. Phone and WebSocket calls still pay the full active rate while idle, so teams using those connection types will not see the saving unless they switch to WebRTC.
Researched with AI assistance and reviewed for accuracy.
Follow on WhatsApp
Follow Daily Tech & AI on WhatsApp for the latest tech news, AI updates and research breakthroughs: https://www.whatsapp.com/channel/0029Vb9XzdHDJ6H4TFNAl61r
Sources
- https://developers.openai.com/api/docs/models/gpt-live-1.1
- https://developers.openai.com/api/docs/pricing?multimodal-image-pricing=batch
- Internal link: https://amjadphd.blogspot.com/2026/11/signal-blocks-ai-agents-private-chats.html
- Internal link: https://amjadphd.blogspot.com/2026/11/gemini-computer-use-mac.html
Comments
Post a Comment