Voice
Standard vs Realtime Voice
Compare Standard Mode (VAD → STT → LLM → TTS) and Realtime Mode (Gemini Live, xAI Grok) — choose the right voice mode.
Standard Mode
Loading diagram…
Realtime Mode
Loading diagram…
Comparison
| Feature | Standard Mode | Realtime Mode |
|---|---|---|
| Architecture | Chained speech-to-text → AI → text-to-speech | Direct speech-to-speech |
| Latency | Moderate (processes in stages) | Low (instant speech-to-speech) |
| Processing model | Standard pipeline: ASR + LLM + TTS | Separate realtime voice model |
| Voice Quality | Very good (dedicated TTS) | Excellent (native audio) |
| Tool Support | Full (web search, KB, etc.) | Limited (provider-specific) |
| Text Input | LLM → TTS pipeline | Sent directly to realtime session |
| Plan Required | Any plan with voice | Plan with Realtime models |
When to Use Standard
- Customer support — Structured Q&A with knowledge base
- Flexible LLM choice — Standard uses a separate LLM for voice turns and a separate TTS model for spoken output
- Tool usage — Needs web search, image generation, etc.
- Cost control — Usually cheaper than Realtime for the same task
When to Use Realtime
- Natural conversations — Back-and-forth dialogue
- Low latency — Instant responses feel more natural
- Simple queries — Don't need complex tool integrations
- Voice-first experience — Prioritizes live conversation feel over long-form answers
Voice usage consumes credits based on the model and usage. See Credits & Usage for calculation details, and Plans & Pricing for Realtime availability.