GPT-Live (ChatGPT Voice)
A Tier · 8.6/10
OpenAI's full-duplex voice models (launched 2026-07-08) now powering ChatGPT Voice -- listens and speaks at the same time, backchannels naturally, and delegates hard questions to GPT-5.5 in the background while keeping the conversation going. GPT-Live-1 for paid tiers, GPT-Live-1 mini for Free
Score Breakdown
The Good and the Bad
What we like
- +Full-duplex architecture is a real generational change -- it processes input while generating output, so it can backchannel ('mhmm'), handle interruptions, wait while you think, and even do live translation instead of rigid turn-taking
- +Delegation is the clever part: hard questions route to GPT-5.5 (Instant or Thinking) in the background while the voice keeps talking -- frontier-model answers without killing conversational flow
- +The benchmark deltas over Advanced Voice Mode are enormous: GPQA 84.2% vs 45.3%, BrowseComp (agentic search) 75.2% vs 0.7% at high effort; humans preferred it over AVM in ~70-76% of matched conversations
- +Visual answer cards (weather, stocks, sports, maps) render while you talk -- voice plus glanceable UI beats voice alone for a lot of everyday queries
- +Voice-specific safety was built in from day one: real-time steering of unsafe output, adapted crisis-support flows, teen protections with parental notifications
What could be better
- −No API at launch -- developers can't build on GPT-Live yet (sign-up form only), so it's a ChatGPT feature today, not a platform
- −No video or screen-share support at launch -- you have to fall back to legacy Standard/Advanced Voice Mode for those
- −Predefined voices only (nine, remastered) -- deliberately no voice cloning, so ElevenLabs/Grok Voice keep that use case
- −Optimized for ChatGPT's most popular languages; OpenAI concedes non-native accents and fluency gaps in others
- −Delegation quality depends on the background model tier you can access -- Free users on mini get the smallest slice of the experience
Pricing
Included in ChatGPT (no separate price)
- ✓GPT-Live-1: default ChatGPT Voice model for Go, Plus, and Pro users
- ✓GPT-Live-1 mini: default for Free users
- ✓Reasoning-effort choice per conversation: Instant (GPT-5.5 Instant behind the scenes) or Medium/High (GPT-5.5 Thinking)
- ✓Rolling out globally 2026-07-08 across iOS, Android, and chatgpt.com
- ✓No API at launch -- OpenAI says API access is 'coming soon' (sign-up form for developers/enterprises)
Known Issues
- LAUNCH (2026-07-08): OpenAI launched **GPT-Live**, a new generation of full-duplex voice models that replaces the default ChatGPT Voice experience (150M+ weekly voice users). **GPT-Live-1** becomes the default for Go/Plus/Pro; **GPT-Live-1 mini** for Free; rolling out globally across iOS/Android/web. Architecture: continuous listen-while-speaking (interaction decisions many times per second) + **delegation** -- search/reasoning/agentic work is handed to GPT-5.5 (Instant, or Thinking at Medium/High effort) in the background while the conversation continues; OpenAI says the background model will track its latest frontier releases. Vendor benchmarks vs Advanced Voice Mode: **GPQA 76.5-84.2% (by effort) vs 45.3%; BrowseComp 35.1-75.2% vs 0.7%**; preferred over AVM in 75.7% (Live-1) / 69.2% (mini) of matched 5-10 min conversations. Also: visual answer cards (weather/stocks/sports/maps), nine remastered predefined voices (no cloning by design), voice-adapted safety stack (real-time steering, crisis flows, teen protections + parental notifications, system card at deploymentsafety.openai.com/gpt-live). Launch limitations: **no API yet** ('coming soon,' sign-up form live), **no video/screen-share** (legacy Standard/Advanced Voice Mode remain available for those), uneven fluency outside major languagesSource: OpenAI (openai.com/index/introducing-gpt-live/), TechCrunch, MacRumors · 2026-07-08
Best for
Anyone who talks to ChatGPT -- commute Q&A, language practice, hands-free help, kids' stories. It's the new default, free with every tier, and the conversational feel is the best shipping voice AI experience right now.
Not for
Developers (no API yet), anyone needing voice cloning or custom voices (ElevenLabs, Grok Voice), video/screen-share workflows (stay on legacy Advanced Voice Mode until GPT-Live adds it), or heavy non-English users in less-supported languages.
Our Verdict
GPT-Live is the biggest step-change in consumer voice AI since Advanced Voice Mode shipped: full-duplex conversation that finally stops feeling like walkie-talkie turns, plus a delegation trick that quietly gives your voice assistant frontier-model intelligence without breaking flow. The GPQA and BrowseComp deltas over AVM are not incremental -- they're a different class. What keeps it from a higher score is scope: no API, no video/screen-share, no custom voices. As a ChatGPT feature it's excellent and free; as a platform it hasn't started yet. If the API lands at reasonable pricing, this becomes a direct threat to every voice-agent stack.
Sources
- OpenAI: Introducing GPT-Live (2026-07-08) (accessed 2026-07-09)
- OpenAI GPT-Live system card (accessed 2026-07-09)
- OpenAI: GPT-Live API notification form (accessed 2026-07-09)
Explore more GPT-Live (ChatGPT Voice) rankings
Deeper leaderboards, benchmarks, task-specific tier lists, and status/pricing pages for GPT-Live (ChatGPT Voice).
The Tier List Tuesday
Weekly newsletter: tier movers, new entrants, and the VS of the week. Built from our daily AI-tool sweeps. No spam, unsubscribe anytime.
Alternatives to GPT-Live (ChatGPT Voice)
ElevenLabs
Best-in-class AI voice generation -- now includes 11.ai (MCP-based voice assistant), Eleven v3 expressive speech, and IBM watsonx partnership. $500M raise at $11B valuation (Feb 2026)
Murf AI
Text-to-speech that actually sounds like a real person read your script -- not a robot trying its best
Descript
Edit audio and video by editing text -- the 'Google Docs of media editing' actually lives up to the hype
Speechify
Text-to-speech reader that turns articles, docs, and PDFs into natural-sounding audio
Microsoft MAI-Voice-2
Microsoft's in-house expressive TTS model -- MAI-Voice-2 launched 2026-06-02 at Build: 15 languages (up from English-only), granular emotion-tag control, zero-shot voice cloning from a 5-60s clip, and preferred over MAI-Voice-1 72% of the time. In speaker-similarity tests its speech is 'indistinguishable' from real recordings. On Azure Foundry + integrated into VS Code and Dynamics 365 Contact Center; lower-cost MAI-Voice-2-Flash coming. Original MAI-Voice-1 shipped 2026-04-02
Grok Speech (STT + TTS APIs)
xAI's standalone voice APIs -- launched 2026-04-17. Built on the stack that powers Grok Voice, Tesla vehicles, and Starlink customer support. $0.10/hr STT batch, $4.20 per 1M characters TTS, 25+ languages. Now 26 flagship TTS voices (21 new added 2026-07-06), custom voice cloning from ~1 min of audio, and a no-code Grok Voice Agent Builder
Cohere Transcribe
Cohere's first audio model -- launched 2026-03-26 under Apache 2.0, 2B parameters, #1 on Hugging Face Open ASR Leaderboard (5.42 avg WER), 14 enterprise-critical languages. Free API with rate limits; Model Vault for production