Best models for real-time voice agents
Build phone-style voice assistants with live audio models billed per minute.
Ordered from lowest price
Live TokenRex prices. Per 1M tokens for chat and embeddings; per unit for media.
| # | Model | Input | Output / unit | Web search / 1K |
|---|---|---|---|---|
| 1 | GPT Live 1 gpt-live-1 | — | $0.00083 / per second | — |
| 2 | GPT-4o Mini Transcribe gpt-4o-mini-transcribe | — | $0.00302 / per minute | — |
| 3 | GPT Transcribe gpt-transcribe | — | $0.00458 / per audio minute | — |
| 4 | Gemini 3.5 Transcribe gemini-3.5-transcribe | — | $0.005 / per minute | — |
| 5 | GPT-4o Transcribe gpt-4o-transcribe | — | $0.00604 / per minute | — |
| 6 | GPT-4o Mini TTS gpt-4o-mini-tts | — | $0.015 / per 1k chars | — |
| 7 | Gemini 2.5 Flash Lite TTS gemini-2.5-flash-lite-preview-tts | — | $0.0171 / per 1k chars | — |
| 8 | Gemini 2.5 Flash TTS gemini-2.5-flash-tts | — | $0.0171 / per 1k chars | — |
| 9 | Gemini 2.5 Pro TTS gemini-2.5-pro-tts | — | $0.0341 / per 1k chars | — |
| 10 | Gemini 3.1 Flash TTS gemini-3.1-flash-tts-preview | — | $0.0341 / per 1k chars | — |
How to choose
- GPT Live handles audio in and out in one connection.
- Combine transcription, a chat model and speech for full control.
Try it with one API key
Use the endpoint for the model’s actual type, not Chat Completions. Images, video, speech, embeddings and live audio have different request formats. Open the model page and Docs before submitting your first request.
Head-to-head comparisons
Questions
- How are these voice models ranked?
- Models are listed from the live TokenRex catalog and ordered by current price. We do not publish invented quality scores; open a model page for source-linked benchmarks where they exist.
- Can I switch models later?
- Yes. Every model uses the same TokenRex API key and endpoint, so switching is a one-line change to the model ID.
- Do prices include a markup?
- Prices shown are the live TokenRex rates billed from your wallet. They update automatically when provider prices change.