GPT-Live-1
A new generation of full-duplex voice models for natural human-AI conversation
Context window
128000
Input / 1M tokens
Not applicable
Output / 1M tokens
Not applicable
Provider
OpenAI
Data verified 2026-09-30
GPT-Live is a full-duplex voice model that listens and speaks simultaneously, enabling natural back-and-forth conversation. It delegates complex reasoning, search, and agentic tasks to GPT-5.5 in the background while maintaining conversational flow. Available in two variants: GPT-Live-1 for paid users and GPT-Live-1 mini for free users.
Capability index
Relative estimates (0-100) to place this model against its peers, grounded in published benchmarks.
How to access it
Rolling out globally to ChatGPT users on Go, Plus, Pro, and Free plans. API access coming soon with sign-up available.
Strengths
- ✓Full-duplex architecture enables simultaneous listening and speaking
- ✓Natural conversational flow with acknowledgments like 'mhmm' and 'yeah'
- ✓Delegates complex reasoning to frontier models without breaking conversation
- ✓Supports live translation and interruption handling
- ✓Strongly preferred over Advanced Voice Mode in human evaluations
Best for developers who...
When to choose it (and when not to)
Reach for GPT-Live-1 when...
- →Users seeking natural, conversational voice interaction
- →Applications requiring real-time turn-taking and interruption handling
- →Tasks needing both quick responses and complex reasoning in background
Look elsewhere if...
- ✕Applications requiring video or screen sharing (not yet supported)
- ✕Use cases needing full multilingual parity (optimized for most languages)
How to use it
- ›You can interrupt naturally mid-response
- ›Pause when thinking; the model waits instead of interrupting
- ›Select reasoning levels: Instant, Medium, or High for task complexity
Quickstart
Voice (iOS/Android/Web)Open ChatGPT app → Tap Voice button → Speak naturally to GPT-LiveNo API access at launch; developers can sign up for notification
API model id: gpt-live-1
Benchmarks
| Benchmark | Score | Notes |
|---|---|---|
| GPQA | 84.2% | Tests expert-level scientific reasoning across biology, chemistry, and physics |
| BrowseComp | 75.2% | Tests agentic web search and ability to find difficult-to-locate information |
| Human Conversational Preference | Strongly preferred | 5-10 minute matched conversations measuring overall preference, turn-taking, interruptions, flow, and naturalness |
| t³-Voice Telecom | 65 | Internal benchmark for telecom support tasks; GPT-Live-1 High setting |
| τ³-Voice Telecom | Not disclosed | Internal benchmark on realistic multi-turn telecom support tasks; Advanced Voice Mode ~30% completion |
| Human Preference | 75.7% | Users preferred GPT-Live-1 over Advanced Voice Mode |
| GPQA (expert-level scientific reasoning) | 84.2% | GPT-Live-1 High reasoning tier; Advanced Voice Mode scored 45.3% |
| BrowseComp (agentic web search) | 75.2% | GPT-Live-1; Advanced Voice Mode scored 0.7% |
| Human preference testing | 75.7% | Users preferred GPT-Live-1 over Advanced Voice Mode in 75.7% of comparisons |
| GPQA (expert-level science reasoning) | 84.2% | GPT-Live-1 vs 45.3% for Advanced Voice Mode |
| User preference testing | 75.7% | Users chose GPT-Live-1 over Advanced Voice Mode |
| Tau3 (airline/retail/telecom support) | 83.6% | First-attempt task completion with GPT-6 Astra at medium reasoning effort vs 45.7% for GPT-Realtime-2.1 |
| TauBanking (document retrieval and account-tool use) | 38.1% | With GPT-6 Astra at medium reasoning effort |
| Full Duplex Bench | +30 | Gain over GPT-Realtime-2.1 |
| Tau3 (support tasks) | 83.6% | paired with GPT-6 Astra at medium reasoning |
| TauBanking | 38.1% | document retrieval and account-tool use |
| Tau3 (voice task completion) | 83.6% | Paired with GPT-6 Astra at medium reasoning effort; Tau3 covers airline, retail, telecom support |
| TauBanking (voice task completion) | 38.1% | Paired with GPT-6 Astra at medium reasoning effort; tests document retrieval and account-tool use |
| Tau3 | 83.6% task completion | Paired with GPT-6 Astra at medium reasoning effort; measures frontier voice-agent intelligence on end-to-end tasks |
| Tau3 (Voice) Intelligence | 86.2% | Task-success on spoken customer-service (airline, retail, telecom) |
| Artificial Analysis Conversational Dynamics | 97.3% | Pause handling, turn-taking, interruptions, backchannels |
Source: OpenAI Official Announcement
Compare GPT-Live-1
Compare GPT-Live-1 with any other model
Build a comparison →All model comparisons →