Verified API status

Is the GPT-Live API available?

The GPT-Live-1 API is not generally available. OpenAI currently provides a form to be notified when it becomes available. Developers who need production realtime voice should use the documented GPT-Realtime-2.1 family instead.

Last verified:

GPT-Live-1 API: notification only

The official page says to sign up for a notification when GPT-Live-1 becomes available in the API. It does not publish access instructions, a model ID, pricing, or API documentation.

Open the official notification form

Codex Voice does not open the GPT-Live API

The July 23 desktop integration lets eligible users direct Codex tasks through ChatGPT Voice. It does not publish a GPT-Live-1 API model ID, endpoint, pricing contract, or API-key access.

Understand GPT-Live in Codex

Current Realtime API model choices

Prices are per one million tokens and were verified on the official model pages.

ModelText input / cached / outputAudio input / cached / outputContext / max output
gpt-realtime-2.1$4 / $0.40 / $24$32 / $0.40 / $64128k / 32k
gpt-realtime-2.1-mini$0.60 / $0.06 / $2.40$10 / $0.30 / $20128k / 32k

Choose a transport

TransportUse it forPractical rule
WebRTCBrowser and mobile clientsRecommended for direct client audio; issue ephemeral credentials from your server
WebSocketServer-side agents and controlled media pipelinesKeep standard API keys on trusted infrastructure
SIPPhone numbers, contact centers, and telephonyVerify incoming webhooks and design explicit transfer or hang-up tools

A migration-safe path

Ship with a documented model

Use gpt-realtime-2.1 for quality and reasoning, or mini when latency and cost dominate.

Separate model configuration

Keep the model ID, voice, reasoning effort, prompt, tools, and cost limits outside core product logic.

Wait for an actual GPT-Live API contract

Migrate only after OpenAI publishes official access, model IDs, pricing, modalities, limits, and docs.

GPT-Live API FAQ

Does the form grant GPT-Live-1 API access?

No. Its current wording is a notification signup, not an API key, beta enrollment confirmation, or general-availability announcement.

Which Realtime model should I use?

Start with gpt-realtime-2.1 when quality and complex tool use matter. Evaluate gpt-realtime-2.1-mini when lower price and faster responses are more important.

Can the Realtime models accept video?

No. Their official model pages list text, audio, and image input; text and audio output; and no video support.