GPT-Live-1

Spec

Attribute	Value
Developer	OpenAI
Released	2026-07-08 (global rollout July 8-9)
Announced	2026-07-08
Context window	unknown
Pricing	unknown
License	unknown
Availability	ChatGPT on iOS, Android, ChatGPT.com — GPT-Live-1 default for Go/Plus/Pro; mini default for Free
Model family	GPT-Live-1 (full); GPT-Live-1 mini (free tier)
Modality	full-duplex audio (listen + speak simultaneously)
Predecessor	GPT-Realtime-2 (for consumer use); Realtime API remains for developers

Release Date

2026-07-08

Capabilities

• Full-duplex speech: model listens and speaks at the same time; users can interrupt naturally mid-sentence
• Intelligent delegation: background delegation to GPT-5.5 for queries needing web search, deeper reasoning, or computation; result returns into the voice conversation when ready
• Live translation: real-time bilingual interpretation
• Backchannels: natural affirmations ("mhmm", "yeah") during pauses
• Voice safety: explicit anti-impersonation guardrails; system card published

Benchmarks

Not disclosed; qualitative demonstrations only (turn-taking, interruption naturalness)

Use Cases

• Conversational AI / personal assistant voice interaction
• Live translation and interpretation
• Hands-free agentic workflows
• Natural, low-latency voice Q&A with model-backend escalation

Access

• GPT-Live-1 → default voice for Go, Plus, Pro ChatGPT users
• GPT-Live-1 mini → default voice for Free users
• Platforms: iOS, Android, ChatGPT.com
• Developer Realtime API (separate) remains GPT-Realtime-2 based

Compared To

Model	Type	Duplex	Latency
GPT-Live-1	Full-duplex voice	✅	<300ms target
GPT-Realtime-2	Reasoning voice (API)	partial	~600ms
Claude Sonnet 5 voice	—	❌	—

Sources

• OpenAI: Introducing GPT-Live (source)
• TechCrunch
• System Card

Related

• OpenAI
• GPT-Realtime-2 (OpenAI) — predecessor for consumer voice; developer API remains
• Agents (LLM Agents) — full-duplex voice as the conversational agent interface