GPT-Live-1 Pricing & Voice Agent Cost Calculator
GPT-Live-1 voice sessions cost $0.05 per minute, billed per second. That is the front-end voice layer—not a universal all-in agent price. Backend model inference, tools and telephony can add separate charges.
- Voice session: $0.05/minute, billed per second without whole-minute rounding.
- Backend Responses calls use the configured model's normal pricing.
- Tool and external API usage is separate.
- OpenAI documents SIP/telephony integration paths but does not publish one universal carrier fee; enter your provider cost separately.
Voice-agent cost calculator
Estimate monthly workload cost
This calculator separates known GPT-Live session pricing from workload-specific inputs. It does not guess your carrier, backend-model or third-party tool rates.
What GPT-Live-1 is
GPT-Live-1 is OpenAI's full-duplex voice model: it can listen while speaking and handle interruptions more naturally than a strict turn-by-turn pipeline. The live model manages the spoken conversation and can delegate deeper reasoning or tool work to a separate backend.
Where the money goes
| Layer | How it is priced | What changes cost |
|---|---|---|
| GPT-Live voice | $0.05/minute, billed per second | Conversation duration |
| Backend model | Normal selected-model and tool pricing | Delegations, context and output |
| External tools | Your provider's pricing | Searches, CRM calls, bookings and other actions |
| Telephony/transport | Provider-specific | Phone numbers, call minutes, regions and routing |
GPT-Live versus Realtime and chained voice stacks
GPT-Live is designed for a conversation that can continue while a backend agent works. A cascaded speech-to-text → language model → text-to-speech stack gives you independent component choice and pricing, but adds integration boundaries and often more latency. Previous Realtime-style voice agents keep model interaction inside the realtime conversation; GPT-Live adds explicit Responses or client delegation for longer or deeper work.
Choose Responses delegation when OpenAI's managed backend workflow fits. Choose client delegation when your application must control backend context, execution or which results return to the conversation. Compare latency, task success and total cost on your own workload.
Prompt-to-voice: free creation, normal GPT-Live usage charges
Documentation checked October 20, 2026: OpenAI's prompt-to-voice workflow is generally available to API customers. Send a text description to POST /v1/audio/voices with type: "prompt". No reference audio or consent recording is required for this prompt-generated workflow; recording-based custom voices have different requirements.
Creating a voice is free, with up to 100 saved prompt-generated voices per organization. Speech generated with it incurs normal GPT-Live session charges, not a new standalone voice-creation token rate. Prompt-generated voices work with GPT-Live, not the standard Text to Speech API, Realtime API or Chat Completions audio.
Voice creation is not compatible with Zero Data Retention. Check generation eligibility separately before using it in a regulated workflow. Disclose synthetic speech and do not use descriptions to impersonate a real person. This capability does not change the calculator's existing cost components.
Connection and telephony options
- WebRTC for browser voice applications.
- WebSockets for server-side audio integrations.
- Server-side controls for a sideband connection to an existing session.
- SIP for telephony integration. Carrier charges remain separate from GPT-Live pricing.