ChatGPT's Voice Mode Just Got a Major Upgrade: What Is GPT-Live? (Updated September 2026)
ChatGPT Voice launched as a full-duplex upgrade in July 2026 — and in September 2026 gained email, calendar, and Slack access. Here's what changed across both updates, who gets which features, and what it means for how you work.
So — which one should you buy?
September 23, 2026 update. OpenAI expanded ChatGPT Voice with plugin access, including email, calendar, and Slack — and brought the Work tab to mobile. Plus and Pro users can now draft emails, summarize Slack messages, create documents and presentations, and browse the web, all by voice on their phones. Free and Go users get plugin and connected-app access via voice. The full breakdown is in the September 2026 section below.
July 31 update. OpenAI's current desktop documentation says GPT-Live manages the live conversation while GPT-5.6 Terra starts and coordinates tasks in the app. The original July 8 launch description below referred to GPT-5.5 as the background model; that detail has now been superseded.
TL;DR. On July 8, 2026, OpenAI replaced ChatGPT's Advanced Voice Mode with GPT-Live — a fully rebuilt voice system that can listen and speak at the same time. Free users get GPT-Live-1 mini automatically; paid users (Go, Plus, Pro) get the full GPT-Live-1. In September 2026, Voice gained plugin access (email, calendar, Slack) and Work tab support on mobile. Rolling out now on iOS, Android, and ChatGPT.com.
What changed (July 2026)
If you've used ChatGPT's voice mode for anything beyond a quick query, you've probably run into the awkwardness: the model finishes its sentence before registering that you've started talking, or you both pause waiting for the other to go first. That limitation was structural — the old Advanced Voice Mode used a half-duplex system, the audio equivalent of a walkie-talkie. One side speaks; the other waits.
GPT-Live changes this at the architecture level. It's full-duplex, meaning ChatGPT processes your voice input and generates its response simultaneously. In practice:
- You can interrupt naturally. If ChatGPT starts explaining something you already know, you can jump in and redirect it — the model registers your voice mid-sentence rather than after it finishes.
- It signals it's listening. GPT-Live uses brief acknowledgments ("mhmm," "got it") to indicate active processing rather than going silent.
- Pauses feel less awkward. If you trail off while thinking, GPT-Live holds space rather than immediately filling the silence.
OpenAI describes the model as making decisions "many times per second" about whether to speak, pause, interrupt, or invoke a tool. The result is a conversation rhythm that's closer to talking with a person than navigating a voice menu.
Two versions, two tiers
OpenAI launched two GPT-Live models on July 8:
| Model | Who gets it | Notes |
|---|---|---|
| GPT-Live-1 | ChatGPT Go, Plus, Pro (paid) | Full-featured; highest user-preference ratings |
| GPT-Live-1 mini | Free ChatGPT accounts | Replaces Advanced Voice Mode as default |
Both run on iOS, Android, and ChatGPT.com. Both can now handle web searches during voice conversations — something the old Advanced Voice Mode couldn't do at all.
If you're on a paid ChatGPT plan, GPT-Live-1 is already your default voice mode. If you're on the free tier, GPT-Live-1 mini replaces Advanced Voice Mode automatically — no settings change needed.
Web search in voice sessions: the bigger deal
The most significant functional improvement at launch wasn't the conversational flow — it was that GPT-Live can look things up while you're talking.
The old Advanced Voice Mode was isolated from the live web. If you asked for current prices, recent news, or anything requiring a real-time lookup, it had to work from training data alone — so it either guessed or declined.
At launch, GPT-Live handled this by delegating to GPT-5.5 in the background. When a question needed web search, deeper reasoning, or a more complex calculation, GPT-Live handed it off behind the scenes, continued the conversation naturally, and brought the result back when ready. OpenAI's current desktop documentation now identifies GPT-5.6 Terra as the model that starts and coordinates tasks while GPT-Live manages the live conversation.
The benchmark difference was stark. On BrowseComp — which tests web-search task performance — GPT-Live-1 scores 75.2%. The old Advanced Voice Mode scored 0.7%.
For professionals, this opened up the kinds of voice queries that actually matter: checking a policy change, asking for a current rate, getting a quick summary of something recent.
What didn't work at launch
Two limitations worth knowing:
No video or screen sharing. GPT-Live doesn't currently support showing ChatGPT your screen or using the camera during a voice conversation. The legacy ChatGPT Voice mode is still required for that. OpenAI has not given a timeline for when video support will arrive.
Non-English languages are uneven. OpenAI acknowledged that GPT-Live performs inconsistently in non-English languages. The system is optimized for "most spoken languages" but demo sessions at launch showed noticeable issues in Hindi and others.
September 2026: ChatGPT Voice gets email, calendar, and Slack
On September 23, 2026, OpenAI pushed a substantial update to ChatGPT Voice: plugins now work in voice mode for the first time, and the Work tab — previously desktop-only — is now on mobile. The result is a voice assistant that can reach into your inbox, calendar, and Slack, not just answer questions about them.
Three things that changed
1. Plugin access in voice. Plugins now connect to ChatGPT Voice directly. That includes email (Google and Microsoft), calendar, Slack, and other connected apps. You can ask Voice to check what's on your calendar tomorrow, summarize the last 24 hours of Slack messages in a channel, or draft a reply to an email — all without touching your keyboard.
2. GPT-6 model support. Voice sessions can now leverage GPT-6 Astra, Sol, and Luna — OpenAI's three GPT-6 tier models (Astra being the most capable, Luna the most efficient). In Work-tab sessions, this means more complex multi-step tasks are handled by the same models powering ChatGPT's most capable agents.
3. Work tab on mobile. The ChatGPT Work tab, which lets users create documents, presentations, spreadsheets, and web apps via an agent, is now available by voice on mobile. You can start a task on your phone and pick it up on desktop. Voice also now produces richer text output during conversations and makes switching between text and voice easier.
Who gets what
| Plan | What Voice can now do |
|---|---|
| Free | Plugin and connected-app access via voice |
| Go | Plugin and connected-app access via voice |
| Plus | Full Work tab on mobile (email, Slack, docs, presentations, sites, finances); limited GPT-6 Astra allowance |
| Pro | Full Work tab on mobile; full GPT-6 Astra access |
To use the new features, update to the latest version of the ChatGPT app on iOS or Android.
What this means in practice
For professionals who already use ChatGPT voice for commuting, meeting prep, or hands-free work, this is a meaningful step up. The difference between "ask questions" and "get things done" was mostly about app access — Voice could tell you what was on your calendar but couldn't show it to you or book anything. That boundary is gone now.
Some concrete use cases that are now in scope by voice:
- "What meetings do I have after 2pm today?" (calendar lookup)
- "Summarize the last 50 Slack messages in the #launches channel" (Slack plugin)
- "Draft a reply to Sarah's email about the Q3 budget" (email plugin)
- "Create a one-page brief about [topic] and save it as a doc" (Work tab)
The caveat: plugin quality varies by provider, and connected apps require prior authorization in your ChatGPT settings. If you haven't connected email or Slack to ChatGPT yet, you'll need to do that through the plugins directory first.
Sources
- OpenAI: Introducing GPT-Live — official announcement, July 8, 2026
- OpenAI: ChatGPT Voice — current desktop voice documentation
- TechCrunch: ChatGPT mobile app gets voice-based agentic features — September 23, 2026
- The Decoder: ChatGPT Voice gets closer to "Her" with email, calendar, and Slack access — September 23, 2026
- 9to5Mac: OpenAI just upgraded ChatGPT Voice in three ways — September 23, 2026
- TechCrunch: OpenAI releases new voice models for more natural live conversations — July 8, 2026
- The Decoder: ChatGPT can now listen and talk at the same time, making AI conversations seem more human — July 8, 2026
So — which one should you buy?
Set up AI for your job — free, in about 2 minutes
Pick your profession and get your first working AI tool, a step-by-step guide, and a $0 plugin to take home. No credit card.
Get my free setupFrequently asked questions
What is GPT-Live and how is it different from ChatGPT's old voice mode?+
GPT-Live is OpenAI's new voice model for ChatGPT, launched July 8, 2026. The biggest difference: it's full-duplex, meaning it can listen and speak at the same time — the same way humans talk. The old Advanced Voice Mode was half-duplex: it took turns speaking and listening in a rigid back-and-forth. GPT-Live feels more like a real conversation because it can catch an interruption mid-sentence, hold space while you think, and signal it's listening with brief acknowledgments ('mhmm,' 'got it') rather than going silent.
Which ChatGPT plan gets GPT-Live-1 versus GPT-Live-1 mini?+
Free ChatGPT users get GPT-Live-1 mini by default — it replaces Advanced Voice Mode automatically. Paid users on ChatGPT Go, Plus, and Pro plans get the full GPT-Live-1 model. Both versions are available on iOS, Android, and ChatGPT.com.
What does 'full-duplex' mean in plain English?+
A walkie-talkie is half-duplex — one person talks, the other waits, then you switch. A regular phone call is full-duplex — both sides can send and receive at the same time. GPT-Live works like a phone call: it continuously processes your voice while generating its response, so it can stop mid-sentence if you start talking, rather than finishing before 'noticing' you interrupted.
Can ChatGPT Voice now access my email, calendar, and Slack?+
Yes, as of September 23, 2026. ChatGPT Voice can now use plugins, which means it can look up calendar events, draft and send emails, and summarize Slack messages — all by voice. The feature is available across plans: Free and Go users get plugin and connected-app access via voice; Plus and Pro users also get the full Work tab on mobile, which adds document creation, presentations, site building, and finance tasks.
Which ChatGPT plan do I need for voice-based email and Slack access?+
Basic plugin and connected-app access via voice is available on all plans, including Free, as of September 2026. For the full Work tab on mobile — drafting emails, summarizing Slack, creating documents and presentations, building sites, and accessing finances — you need Plus or Pro. Pro gets full GPT-6 Astra access in Work; Plus gets a limited Astra allowance.
Can ChatGPT voice mode still be used with video or screen sharing?+
Not yet with GPT-Live. GPT-Live doesn't currently support video or screen sharing. If you need to show ChatGPT your screen or use the camera during a voice conversation, you still need the legacy ChatGPT Voice mode. OpenAI hasn't announced a timeline for when video support will come to GPT-Live.
How does GPT-Live handle questions that need web search or complex reasoning?+
GPT-Live manages the real-time conversation while GPT-5.6 Terra starts and coordinates tasks in the background. At the July 8 launch, OpenAI described GPT-Live delegating deeper work to GPT-5.5; that coordinator has since moved to the GPT-5.6 family. In Work-tab voice sessions, GPT-6 Astra handles the more complex task work. The old Advanced Voice Mode couldn't do web search during a voice session at all.
Related Guides
Can AI Fire You in California? What the New Workplace AI Laws Mean (October 2026)
On September 30, 2026, Governor Newsom signed California's No Robo Bosses Act (SB 947) and three related workplace AI laws. Starting July 1, 2027, an employer can't fire or discipline you based on an AI system alone, and must tell you when AI was used. Here's what each law does, when it starts, and what to do if you work in or manage people in California.
Should You Give an AI App Full Disk Access on Your Mac? (October 2026)
On October 2, 2026, Apple said it will make Full Disk Access on macOS harder to grant because AI agents make the permission riskier. Here's what Full Disk Access actually lets an app read, why Apple is tightening it, how to check which apps have it today, and when it's reasonable to say yes.
What Is Gemini 4 Argon, and Can You Use It Yet? (October 2026)
Google announced Gemini 4 Argon, its most capable model, on September 30, 2026, but released it only to vetted cybersecurity defenders. It is not in the Gemini app or the public API yet. Here's what Argon is, why Google is holding it back, who gets it next, and what to do in the meantime.