Tasks and inputs
GPT-Live-1.1 is listed for Audio-to-audio, Text → Speech, Speech → Text. It accepts audio, text and produces audio, text. Use the publisher’s input format and preprocessing for your chosen deployment.
GPT-Live-1.1 is OpenAI's full-duplex Live API voice model, with more accurate transcription and steadier multilingual accents than GPT-Live-1.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Indicative: $0.05 per minute active, $0.02 per minute standby (WebRTC sessions with inference paused), billed per second; backend model and tool usage billed separately. Confirm on your own OpenAI account.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
If your voice agent already runs on GPT-Live-1, the upgrade is a one-line change. GPT-Live-1.1, released on the API on 19 November 2026, slots into an existing Live session configuration: swap the model string to gpt-live-1.1 and keep the rest of your setup. The interesting question is whether the change is worth making, and for whom.
The short answer: teams whose callers do not all speak the same language. OpenAI's stated improvements are narrow and specific. The new model transcribes speech more accurately, pronounces words better, and holds a more consistent accent and voice when a conversation moves between languages. Nothing about the architecture or the price changes.
The design inherited from GPT-Live-1 is still the point. One model handles the conversation itself, full duplex, so it keeps listening while it talks and copes with interruptions. You describe in plain language when it should hand a job to a backend, and that backend does the reasoning and calls the tools.
Two delegation modes exist. With Responses delegation, OpenAI runs a supported Responses model behind the voice layer and passes context and results back and forth; your app still executes its own function tools. With client delegation you plug in any model, agent harness or service and own its context entirely. The mode is fixed when the session is created. Switching means a new session.
| Spec | GPT-Live-1.1 |
|---|---|
| API ID | gpt-live-1.1 (single snapshot, also the default) |
| Released | 19 Nov 2026 |
| Endpoint | Live API only, v1/live/sessions |
| Input / output | Audio and text in, audio and text out |
| Features | Streaming, function calling |
| Transports | WebRTC (recommended), WebSockets, SIP telephony, server-side sideband |
| Price, active | $0.05 per minute, billed per second |
| Price, standby | $0.02 per minute (WebRTC, inference paused) |
It is not offered on Chat Completions, Responses, the Realtime endpoints, Batch or fine-tuning. Backend model and tool usage are billed separately from the voice minutes, so a cheap session can still produce an expensive bill if the delegated model is heavy. Treat the per-minute rates as indicative and check them on your own account.
The same release added a standby setting for WebRTC sessions on both GPT-Live versions. When nobody is speaking, inference pauses; it resumes when someone talks again. Paused time bills at $0.02 a minute instead of $0.05.
For a support line where most of the call is the customer reading out an order number, that matters little. For a kiosk, an in-car assistant or a companion app left open for an hour, it can more than halve the voice bill. Standby is not exclusive to 1.1, though, so it is no reason on its own to upgrade.
Probably, if transcription errors are your top complaint. Here is a quick way to find out:
One operational detail: the two versions have separate concurrent-session limits, each applied across inference regions. A staged rollout can therefore run both side by side without one eating the other's headroom.
OpenAI's model overview now lists only the 1.1 release, but the original is not on the deprecations page and still has its own price row. Existing deployments keep working. If you need a classic speech-to-speech pipeline rather than a delegating voice layer, compare GPT-Realtime-2.1, or browse other audio-to-audio models and the full OpenAI lineup.
Compare the model on the work you need to complete.
GPT-Live-1.1 is listed for Audio-to-audio, Text → Speech, Speech → Text. It accepts audio, text and produces audio, text. Use the publisher’s input format and preprocessing for your chosen deployment.
Compare pronunciation, voice consistency and time to the first playable audio. Check the service’s billing unit for the actual script.
Use the model identifier and account requirements for the selected API route. Confirm its request limits and output format with one complete task before connecting a larger workflow.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
Microsoft
MAI Realtime 1 is Microsoft AI's first speech-to-speech model for live voice agents, from $0.01 per audio minute, in limited preview via MAI Labs and Foundry.
rime
Rime Prism is a speech-to-speech model for enterprise voice agents that settles each reply as inspectable text first, with parallel tool calls.
Liquid AI
LFM3-Audio-2B is Liquid AI's open 2B speech model: ASR, TTS, voice chat and interleaved text-and-speech output from one local checkpoint.
Alibaba
Qwen-Audio-3.1-Realtime Flash is Alibaba's lighter full-duplex speech model with audio and text in and out, function calling, web search and a 262K context.
ElevenLabs
Eleven Realtime v1 Max is ElevenLabs' reasoning speech-to-speech agent for support and multi-step tool workflows, OpenAI Realtime-compatible.
ElevenLabs
Eleven Realtime v1 Mini is ElevenLabs' fast, low-cost speech-to-speech voice agent stack, with an OpenAI Realtime-compatible API and 90+ languages.