OpenAI now has two voice technologies with names one letter apart in spirit, and the July 8 GPT-Live announcement made the confusion mainstream. Search for “GPT Live API” and you’ll land on GPT-Realtime documentation; read launch coverage and you’ll find the two used interchangeably. They are not the same thing.
The one-line disambiguation: GPT-Live is the consumer voice experience inside ChatGPT. GPT-Realtime is the developer API family for building your own voice applications. Here’s the full picture, and how to tell which one you’re looking for.
The two stacks, side by side
| GPT-Live | GPT-Realtime | |
|---|---|---|
| What it is | Voice model family powering ChatGPT Voice | Speech-to-speech models in the Realtime API |
| Who it’s for | ChatGPT users | Developers building voice products |
| Where it runs | ChatGPT apps (iOS, Android, web), CarPlay | Your application, via API |
| API access | None yet (“soon,” per OpenAI) | Generally available |
| Current models | GPT-Live-1, 1 mini, 1 Medium, 1 High | gpt-realtime, gpt-realtime-1.5, gpt-realtime-2.1, 2.1-mini |
| Architecture | Full-duplex + background delegation to GPT-5.5 | Single speech-to-speech model, fast turn-based |
| Conversation style | Listens while speaking, backchannels | Low-latency turns with interruption handling |
| Extras | Web search, translation, visual cards, memory | Function calling, MCP servers, SIP calling, image input |
| Pricing | Not published (bundled in ChatGPT tiers) | Published: e.g. gpt-realtime at $4/M in, $16/M out |
What GPT-Live is
GPT-Live is the new default for ChatGPT Voice: full-duplex models that process incoming audio while generating output, deciding many times per second whether to speak, listen, pause, or interrupt. When a question needs search or deeper reasoning, GPT-Live “can delegate the task to another model like GPT-5.5” and weave the result back into the conversation.
You don’t integrate GPT-Live. You use it, in the ChatGPT app. Paid tiers default to GPT-Live-1; the Free tier gets GPT-Live-1 mini. The setup guide covers variants and platforms.
What GPT-Realtime is
GPT-Realtime is the model family behind OpenAI’s Realtime API: speech-to-speech models you connect to over WebSocket or WebRTC, with SIP support for phone calling and remote MCP servers for tool use. It’s generally available, versioned, and priced like an API product. The newest members, gpt-realtime-2.1 and 2.1-mini, shipped on July 6, 2026, two days before GPT-Live was announced.
This is the stack our hands-on guides cover: the original gpt-realtime walkthrough, GPT-Realtime-2, and GPT-Realtime-2.1-mini. If you’re building with it, Apidog can drive the WebSocket sessions directly, which turns opaque realtime debugging into inspectable event streams; download it free to test a session against gpt-realtime-2.1 in minutes.
Why the confusion is understandable
The two stacks are converging from opposite directions. GPT-Realtime brought speech-to-speech latency low enough for production agents; GPT-Live added the full-duplex layer and delegation on top, but only inside ChatGPT. Architecturally, GPT-Live looks like what the Realtime API’s next generation wants to be, and OpenAI has said the GPT-Live models are coming to the API “soon.”
Until that happens, the practical mapping is:
- “I want to talk to an AI.” GPT-Live, in ChatGPT. No code involved.
- “I want my app or phone line to talk to users.” GPT-Realtime, via the Realtime API, today.
- “I want GPT-Live’s full-duplex behavior in my app.” Not available yet. The closest approximations, including hand-rolling the delegation pattern, are in Is there a GPT-Live API?.
What happens when GPT-Live reaches the API
OpenAI hasn’t said whether GPT-Live models will join the Realtime API, replace it, or arrive as a separate surface. The four-variant structure (Instant-backed and GPT-5.5-Thinking-backed at two effort levels) suggests session-level latency/depth trade-offs. For teams building now, the safe posture is event-driven session code on the Realtime API with delegation logic kept loosely coupled; that ports cleanly under any of the three outcomes.
FAQ
Is GPT-Live the same as GPT-Realtime? No. GPT-Live powers ChatGPT Voice for consumers; GPT-Realtime is the developer model family in the Realtime API.
Can I use GPT-Live in my own app? Not yet. OpenAI plans API availability “soon.” The Realtime API with gpt-realtime-2.1 is the shipping alternative.
Is GPT-Realtime being replaced by GPT-Live? OpenAI hasn’t announced that. GPT-Realtime remains generally available and was updated two days before GPT-Live launched, so treat coexistence as the planning assumption.
Which is more capable? Different jobs. GPT-Live has full-duplex conversation and GPT-5.5 delegation inside ChatGPT; GPT-Realtime has the developer surface: function calling, MCP, SIP, and published pricing.
