Browser or mobile voice agent
Use GPT-Realtime-2.1 over WebRTC, with ephemeral client credentials and server-owned business logic.
Developer status
When will the GPT-Live API be released?
OpenAI has not published a GPT-Live API release date. The official GPT-Live-1 API page is a notification form, not documentation or generally available access.
Last verified:
The direct answer
There is no public GPT-Live-1 model ID, endpoint, price, SDK contract, or release date as of July 28, 2026. Register for the official notification and build current voice products with GPT-Realtime-2.1 instead.
Compare GPT-Live and RealtimeRelease-signal checklist
| Signal | Current evidence | Conclusion |
|---|---|---|
| Official API form | Collects notification signups | Interest registration, not access |
| Model catalog | Lists GPT-Realtime models, not GPT-Live-1 | No public GPT-Live model contract |
| Pricing page | No documented GPT-Live token or minute price | Do not estimate production cost |
| API docs and SDKs | No GPT-Live endpoint examples | Do not invent request syntax |
| Launch post | Says API is planned | Direction exists, date does not |
What developers can do now
Server-managed voice workflow
Use WebSocket when your server owns the long-lived event stream, tools, policy, and observability.
Phone agent
Use SIP for telephony, then design consent, transfers, webhook verification, escalation, and failure recovery.
GPT-Live vs current Realtime API
ChatGPT product behavior does not define an API contract.
| Question | GPT-Live-1 | GPT-Realtime-2.1 |
|---|---|---|
| Generally available API? | No | Yes |
| Public model ID? | No | gpt-realtime-2.1 |
| Public pricing? | No | Published on the model page |
| Connection methods? | Not documented | WebRTC, WebSocket, SIP |
| Migration date? | Not announced | Available now |
Avoid stale or false release claims
A ChatGPT launch, a waitlist form, a social post, or a model name found in client code does not prove general API availability. Require an official model page, pricing, endpoint documentation, and account access before changing production claims.
If a third-party article gives a date, check whether it links to an actual OpenAI announcement. Phrase forecasts as forecasts and keep them out of metadata, FAQ answers, and migration commitments.
Keep your voice architecture model-agnostic: isolate the model ID, transport, session settings, tools, prompts, safety policy, transcript storage, and cost controls. That makes a future GPT-Live migration a measured adapter change rather than a rewrite.
GPT-Live API release FAQ
Is the GPT-Live API available now?
No generally available GPT-Live-1 API contract is documented. The official page is a notification signup.
What is the GPT-Live API release date?
OpenAI has not published a release date.
Can I use GPT-Live-1 through the Realtime endpoint?
No public GPT-Live-1 model ID is documented. Use a listed GPT-Realtime model rather than guessing an identifier.
What should I build with today?
Use GPT-Realtime-2.1 for quality or GPT-Realtime-2.1 mini for a lower-cost voice-agent path, then choose WebRTC, WebSocket, or SIP for the deployment shape.