GPT-Live (ChatGPT Voice)
A Tier · 8.6/10
OpenAI's full-duplex voice models (launched in ChatGPT 2026-07-08) -- listens and speaks at the same time, backchannels naturally, and delegates hard questions to a backend model while keeping the conversation going. **GPT-Live-1 reached the API on 2026-09-10 at $0.05 per minute** for the voice layer, with telephony support, keyword biasing, delegation to any backend model (GPT-6 Astra, Luna, or a third-party model) and a wider voice set. In ChatGPT: GPT-Live-1 for paid tiers, GPT-Live-1 mini for Free. **Five days after the API launch, Google priced Gemini 3.8 Live at $0.005/min audio in and $0.018/min audio out (2026-09-15)** -- a fraction of GPT-Live-1's $0.05/min voice layer, though each stack bills backend or text tokens on top
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
- −The API arrived two months after the consumer launch (2026-09-10) and the $0.05/minute rate is only the voice layer -- every delegated reasoning or tool call bills the backend model on top, so real per-minute cost depends heavily on how much you route to Astra-class models
- −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 the July launch -- resolved 2026-09-10, see the API plan below
API -- GPT-Live-1 (since 2026-09-10)
- ✓Full-duplex voice model in the API: listens and speaks simultaneously, handles interruptions, backchannels and silence in one model
- ✓Delegates reasoning and tool calls to a backend text model you choose -- GPT-6 Astra, Luna, or a third-party model -- so backend token costs are billed separately on top of the per-minute rate
- ✓Telephony support for phone-call agents; native ASR transcripts + response text; alphanumeric understanding and keyword biasing; native turn detection for apps built around turn boundaries
- ✓Tone, pace and style steerable via the system prompt; expanded voice set across accents, dialects and languages, with more promised over the coming months
- ✓Custom voice access is by application through sales, not self-serve
- ✓Also sold packaged as OpenAI Presence (enterprise voice agents) through account directors
Known Issues
- GEMINI 3.8 LIVE UNDERCUTS THE GPT-LIVE-1 API ON THE VOICE LAYER, FIVE DAYS AFTER IT SHIPPED (2026-09-15, competitor vendor-primary): Google launched **Gemini 3.8 Live and 3.8 Live Extended Thinking** in the Gemini API at **$0.005 per minute of audio in and $0.018 per minute of audio out** (Google's own per-minute conversion of $3 and $12 per 1M audio tokens; text $0.75/$4.50 per 1M; free tier available). GPT-Live-1's API rate is **$0.05 per minute** for the voice layer with delegated backend calls billed separately. **This is not a like-for-like comparison, and the difference matters both ways:** Gemini's audio-out price includes its thinking tokens but text, image and tool tokens are billed on top, while GPT-Live-1's per-minute price covers the full-duplex layer only and every delegated reasoning call is a second bill. On a plain conversational minute with little delegation, Gemini's audio-in is a tenth of GPT-Live-1's flat rate and its audio-out well under half; on delegation-heavy workloads both stacks add a backend line and the gap narrows to whatever the backend costs. **On capability claims, Google published third-party leaderboard positions (#1 on Artificial Analysis' Speech to Speech Quality Index at 82.6 for Extended Thinking, second in Speech Agent Arena for 3.8 Live, 68.6% tau-Voice) where OpenAI published its own Full Duplex Bench delta**; neither vendor has published a head-to-head, and the absolute Tau3/Full Duplex numbers this page has been waiting on for GPT-Live-1 are still missing. Google also lists asynchronous function calling (tools run while audio keeps streaming) and 97-language mid-call switching, the two features closest to GPT-Live's delegation pitch. **Practical read: the API price-to-beat for a full-duplex voice agent is now Google's, and OpenAI's $0.05/min has to be justified on delegation quality and the ChatGPT install base rather than on rate.**Source: Google developers post (blog.google/innovation-and-ai/technology/developers-tools/build-real-time-voice-applications-gemini-audio/, 'Sep 15, 2026') and Gemini API pricing page (ai.google.dev/gemini-api/docs/pricing, 3.8 Live row, rendered via Firecrawl); OpenAI GPT-Live-1 API post (2026-09-10, already on this page) -- all fetched 2026-09-17 · 2026-09-15
- GPT-LIVE-1 REACHES THE API -- $0.05 PER MINUTE FOR THE VOICE LAYER, BACKEND BILLED SEPARATELY, AND THE 'COMING SOON' THIS PAGE HAS CARRIED SINCE JULY IS CLOSED (2026-09-10, vendor-primary): OpenAI launched **GPT-Live-1 in the API**, bringing the full-duplex model that replaced ChatGPT Voice on 7/08 to developers. **What is new for the API version rather than a port:** (1) **delegation is open** -- GPT-Live-1 'can delegate reasoning and tool calls to a backend text model like GPT-6 Astra **or a third-party model**', so the voice layer is decoupled from OpenAI's reasoning models (OpenAI's own example pairs it with Luna for scheduling/order updates and Astra for complex issues); (2) **telephony support** for phone-call agents; (3) tone, pace and style steerable via the system prompt; (4) native ASR transcripts and response text, alphanumeric understanding, keyword biasing; (5) native **turn detection** despite not being turn-based, so existing turn-boundary architectures still work; (6) a **wider voice set** across accents, dialects and languages, expanding 'over the coming months'; (7) custom voices by application through sales. **PRICING: $0.05 per minute for the front-end voice layer.** Read that carefully -- it excludes the backend: 'Pair it with the backend model and agent harness that fit your product', which means every delegated reasoning call bills the backend at that model's token rates on top of the per-minute charge. There is no bundled all-in rate. **VENDOR BENCHMARKS (first-party):** Full Duplex Bench **+30 percentage points over GPT-Realtime-2.1**; paired with GPT-6 Astra at medium effort it 'ranks #1 on Tau3' (a voice-agent task suite spanning airline, retail, telecom and banking); other evals ran with Terra (low) as the backend. No absolute scores are given in text, only the chart. **CUSTOMER CLAIMS, ALL VENDOR-SELECTED:** Speak says it cut interruptions during thinking pauses by 'almost 80%' vs turn-based systems; an unnamed healthcare CTO says the cascaded build shrank by 80% and 23K lines; Yelp Host reports better call-handling rates; Intercom's Fin and Cognition's Devin are quoted as integrations. **WHY THIS MATTERS FOR THE COMPETITIVE MAP:** this is OpenAI selling directly into the cascaded STT -> LLM -> TTS voice-agent market that ElevenLabs Agents, Deepgram and the realtime-API integrators occupy, with a single-model pitch on latency and interruption handling. **WHAT DID NOT CHANGE:** the consumer ChatGPT product still lacks video/screen-share and custom voices; the July page text saying 'no API' is superseded by this entry and has been corrected in tagline, pricing, cons, notFor and verdict.Source: OpenAI (openai.com/index/introducing-gpt-live-1-in-the-api/, publicationDateText 'September 10, 2026'; checked for 'Originally published on' / 'Update on' blocks -- none, genuine new post) -- fetched 2026-09-14 via curl with browser UA · 2026-09-10
- A SECOND TRANSCRIPTION DEPRECATION WAVE -- WHISPER-1 AND THE ENTIRE GPT-4o-TRANSCRIBE FAMILY GO ON 2027-02-26 (announced 2026-08-26, vendor-primary): five weeks after the 7/20 legacy-audio notice below, OpenAI opened a **separate, larger** deprecation. Verbatim: '**On August 26, 2026, we notified developers using `whisper-1`, `gpt-4o-transcribe`, `gpt-4o-mini-transcribe`, and `gpt-4o-transcribe-diarize` of their deprecation and removal from the API on February 26, 2027.**' **Replacements for all four are `gpt-live-transcribe` or `gpt-transcribe`.** **THE HEADLINE IS WHISPER.** `whisper-1` has been OpenAI's speech-to-text endpoint since 2023 and is embedded in an enormous amount of third-party tooling, much of it written by people who have not looked at OpenAI's deprecation ledger since. **This is the retirement of the default answer to 'how do I transcribe audio with OpenAI'**, and it gets six months' notice. Note the open-weight Whisper models on Hugging Face are a separate artefact and are not affected by an API deprecation -- **what is ending is the hosted `whisper-1` endpoint**, and self-hosted Whisper keeps working. **READ THE TWO WAVES TOGETHER, BECAUSE THEY OVERLAP CONFUSINGLY:** the 7/20 notice below retires the legacy realtime/audio families on **2027-01-20** and moves `gpt-4o-mini-transcribe-2025-03-20` to a **December 2025 snapshot of the same model** -- and this 8/26 notice then retires that whole `gpt-4o-*-transcribe` line outright on **2027-02-26**. **So if you migrated in July onto a gpt-4o-mini-transcribe snapshot, you migrated onto something that has now itself been deprecated, and you get five extra weeks for the trouble.** The durable target is the `gpt-transcribe` / `gpt-live-transcribe` pair, not any `gpt-4o` transcription ID. **This was announced two days before our 8/28 sweep and we missed it; caught on the 8/31 pass by reading the deprecations doc rather than the newsroom** -- deprecations frequently never get a blog post at all.Source: OpenAI (platform.openai.com/docs/deprecations, fetched as .md -- section '2026-08-26: Transcription models' plus its four-row shutdown table) -- fetched 2026-08-31 via curl · 2026-08-26
- LEGACY AUDIO/REALTIME/TRANSCRIPTION MODELS DEPRECATED (2026-07-20, vendor-primary; removal 2027-01-20): OpenAI notified developers that its legacy audio, realtime, and transcription model families are deprecated and **will be removed from the API on January 20, 2027**. Affected IDs and their replacements: `gpt-realtime` -> `gpt-realtime-2.1`; `gpt-4o-realtime` -> `gpt-realtime-2.1`; `gpt-realtime-mini` / `gpt-4o-mini-realtime` -> `gpt-realtime-2.1-mini`; `gpt-audio`, `gpt-4o-audio`, `gpt-audio-mini`, `gpt-4o-mini-audio` -> `gpt-audio-1.5`; `gpt-4o-mini-transcribe-2025-03-20` -> `gpt-4o-mini-transcribe-2025-12-15`. This is the developer-API voice stack (distinct from the consumer GPT-Live experience below, which has no public API yet). If you build on OpenAI voice/realtime today, plan the migration before the Jan 2027 cutoff -- `gpt-realtime-2.1` / `-mini` are the current GA developer path while the consumer GPT-Live API remains 'coming soon'Source: OpenAI API deprecations (developers.openai.com/api/docs/deprecations) · 2026-07-20
- 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
Anyone needing self-serve voice cloning or custom voices (ElevenLabs, Grok Voice -- OpenAI's custom voices are sales-gated), video/screen-share workflows in ChatGPT (stay on legacy Advanced Voice Mode until GPT-Live adds it), heavy non-English users in less-supported languages, or developers who need a single-model turn-based stack with predictable per-token pricing rather than per-minute voice plus a separately billed backend.
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 kept it from a higher score at launch was scope: no API, no video/screen-share, no custom voices. The API gap closed on 2026-09-10 at $0.05 per minute for the voice layer, with telephony and delegation to any backend model -- which makes it a direct threat to every cascaded STT-LLM-TTS voice-agent stack, and the customer quotes (Yelp, Speak, Intercom's Fin, Cognition) are exactly that market. Still missing in ChatGPT: video/screen-share and custom voices. The new pressure is price: Gemini 3.8 Live (Sept 15) lists at $0.005/min in and $0.018/min out, so OpenAI's $0.05/min now has to earn its premium on delegation quality, not rate.
Sources
- Google Developers: Build real-time voice applications with Gemini 3.8 Live and 3.5 Transcribe -- $0.005/min in, $0.018/min out (2026-09-15) (accessed 2026-09-17)
- Google: Gemini API pricing -- Gemini 3.8 Live / Extended Thinking row (verified 2026-09-17) (accessed 2026-09-17)
- OpenAI: Build more natural voice experiences with GPT-Live-1 in the API -- $0.05/min voice layer, telephony, backend delegation, Full Duplex Bench +30pts vs GPT-Realtime-2.1 (2026-09-10) (accessed 2026-09-14)
- OpenAI deprecations: 2026-08-26 Transcription models -- whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe, gpt-4o-transcribe-diarize removed from the API 2027-02-26 (accessed 2026-08-31)
- OpenAI API deprecations: legacy audio/realtime/transcription removal 2027-01-20 (announced 2026-07-20) (accessed 2026-07-22)
- 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
**ElevenAgents gained two new tools in the week of 2026-10-06: Architect (alpha, 10/06), a built-in expert that diagnoses agent failures from transcripts and drafts tested fixes, and synthetic voice detection (10/09), which flags AI callers in the first seconds of a call -- enterprise-only for now. ElevenCreative added Brand Kit (10/09).** **Two days after the v4 launch, ElevenLabs closed a $300M employee tender at a $22B valuation (2026-09-30), double its February Series D, with enterprise now 55% of revenue -- and the pricing page is running v4 launch promos (extra v4 credits on Creator plans and above, see the Oct 2 note).** **Eleven v4 and Eleven v4 Turbo launched 2026-09-28 -- a fourth-generation voice model on an entirely new architecture**, ranked #1 on Artificial Analysis' Provider Voice Arena and preferred by ~75% of listeners in blind head-to-head tests against Cartesia Sonic 3.6, Inworld TTS-2 and Google's Gemini 3.8 Flash and Flash-Lite TTS; v4 Turbo answers with a ~100ms median inference latency (~150ms to first speech over WebSocket) and is co-optimized with ElevenAgents. Both cover 90+ languages, follow inline audio tags like [laughs] or [phone buzzing] more accurately, clone voices with better speaker similarity, and are live now in ElevenAgents, ElevenCreative and the API (Text to Dialogue). Eleven v3 becomes the previous generation. The 2026-09-10 Universal Music Group licensing deal and CLI v1 (8/24) still stand.
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 line. **MAI-Voice-2.1 and MAI-Voice-2.1-Flash launched 2026-10-01: 23 languages and 26 locales with one voice that keeps the same speaker and a native accent across languages, voice cloning from a few seconds of reference audio with consent guardrails, priced at $22 per 1M characters for 2.1 and $15 per 1M characters for Flash, which generates 45 seconds of audio at 150 ms end-to-end latency and is pitched at voice agents alongside MAI-Transcribe-2-Streaming.** These supersede **MAI-Voice-2 (Build, 2026-06-02)**: 15 languages, emotion tags, zero-shot cloning from a 5-60 s clip, preferred over MAI-Voice-1 72% of the time. Microsoft also shipped a MAI-Voice-2-Flash and MAI-Image-2.5-Pro in July that this page had not recorded -- see the Oct 2 note.
Grok Speech (STT + TTS APIs)
xAI's standalone voice APIs. **Grok Voice Transcribe 2.0 shipped 2026-09-18** -- xAI calls it twice as accurate as 1.0 at the same $0.10/hr batch and $0.20/hr streaming, ranks it first for accuracy among 32 streaming models on Artificial Analysis, and has already made it the default STT model; **1.0 is being retired in the coming weeks, so pin grok-voice-transcribe-1.0 if you need it**. **PRICE CORRECTION: text-to-speech is $15.00 per 1M characters on xAI's live rate card (verified 2026-09-21), not the $4.20 this page carried since April.** Speech-to-speech agent $0.08/min, 26 flagship voices, ~1-minute voice cloning, no-code 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