By Prakash Singh, Founder & CEO, Matrytech Solutions
Short answer: On October 16, 2026, OpenAI made prompt-to-voice generally available for GPT-Live API customers: describe a voice in plain text, save it, and reuse it in live voice sessions. For delivery and logistics businesses this changes the cost of giving a voice agent a consistent, on-brand voice - for delivery-status calls, driver support and dispatch - from a recording and consent process to one line of text. Creating the voice is free; you pay normal session billing when it speaks.

What exactly did OpenAI launch?

Per OpenAI's October 16 release notes: prompt-to-voice is now generally available for API customers. A text description creates a saved voice you can reuse across GPT-Live sessions. OpenAI also added nine new GPT-Live voices covering German, French, Hindi, Japanese and Korean, and (October 15) native Decisions API integration so a live voice session can trigger decisions mid-call as the transcript arrives.
One boundary matters more than the headline: prompt-generated voices work with GPT-Live only - not the separate Text-to-Speech, Realtime or Chat Completions APIs. Voice creation from a recording stays a different, eligibility-limited process with consent requirements. (Source: OpenAI release notes, Oct 15-16 2026; reported via MUVO's Oct 17 daily AI report. Capabilities are provider claims.)

How to use it in practice

  1. Describe the voice in the Playground or API - for example "calm, warm Hindi-English bilingual dispatcher, measured pace" - and save it.
  2. Point your GPT-Live voice session at the saved voice.
  3. Test against your real call scripts before any customer hears it.

The four best use cases in delivery and logistics

  1. Delivery-status ("where is my order") calls with a consistent brand voice instead of a generic bot.
  2. After-hours driver support - Hindi and English are now both first-class, which matters for Indian and Gulf fleets.
  3. Failed-delivery outreach that sounds like your company, not a robot.
  4. Dispatch read-outs to riders who are on the road and cannot read a screen.

What it means if you build a business on it

Voice-agent builds just got cheaper to brand: no studio, no voice-actor pipeline for a prompt-generated voice. The cost model stays usage-based - creation is free, speaking is billed per session - so model your per-call cost before you promise margins. We built a free calculator for exactly that: https://www.matry.tech/voice-ai-cost-calculator/
The compatibility boundary is the trap: prompt voices do not work in the standard TTS or Realtime APIs, so check before you design the architecture. And disclosure rules for AI-voice calls vary by market - get that reviewed for the countries you operate in before launch [general guidance, not legal advice].

Straight answers

Is prompt-to-voice free? Creating the voice is free. Speaking with it is billed as normal GPT-Live session usage (OpenAI, Oct 16).
Can I use it in the standard Text-to-Speech API? No - GPT-Live only, per OpenAI's compatibility note.
Does it support Hindi? Yes - Hindi is one of the nine new GPT-Live voice languages.
Can it clone my own voice? Recording-based voice creation is separate, eligibility-limited, and consent-gated.
Want a voice agent that answers your delivery calls in your brand's voice? Book a free 30-minute call: https://calendly.com/matrytech-info/30min