Compare realtime APIs
Pick up to four APIs from any category and see them side by side.
| OpenAI GPT-Live API OpenAI | Ultravox Realtime Ultravox (formerly Fixie.ai) | Deepgram Voice Agent API Deepgram | |
|---|---|---|---|
| Category | Voice-to-voice | Voice-to-voice | Voice-to-voice |
| Status | GA | GA | GA |
| Est. per minute | $0.05 | $0.05 - 0.055 | $0.041 - 0.163 |
| How that was worked out | Low = voice layer only, $0.05 per minute of session duration with no delegated work. High depends entirely on the backend model, how often the model delegates and tool fees; budget per-minute session cost plus backend token spend measured in your own tests. | $0.05 call rate plus $0.005 SIP when using SIP; LLM and TTS included. | Official tier rates; BYO tiers add your own LLM/TTS bills on top. |
| Pricing model | per-minute | per-minute | per-minute |
| Free tier | Free tier not supported. | 30 free call minutes; playground calls free. | $200 one-time credit for new accounts. |
| Connects by | WebRTC, WebSocket, SIP | WebRTC, WebSocket, SIP, Twilio, Telnyx, Plivo, Exotel | WebSocket |
| Audio in | WebSocket: audio/pcm 24 kHz (default) or 16 kHz mono 16-bit LE, audio/pcmu or audio/pcma 8 kHz; raw bytes base64, even byte length, no container. Format fixed at session start. WebRTC and SIP negotiate codecs (outbound SIP needs Opus + SDES-SRTP). | Raw PCM s16le for server WebSocket (sample rate set in the call medium) | linear16 default at 16 kHz (other encodings/rates configurable) |
| Audio out | Same format as input (one setting covers both). | PCM for server WebSocket; WebRTC handles codecs automatically | linear16 (e.g. 24 kHz), optional container; other encodings configurable |
| Languages | Not listed on the model page. | Multilingual; no list on the pages read. | English by default; multilingual via Nova 'multi' or flux-general-multi with language hints, and multilingual TTS providers. |
| Latency (vendor claim) | Vendor claim: improves Full Duplex Bench score by 30 percentage points over gpt-realtime-2.1 (reported via third-party coverage). No millisecond figure published. | Vendor says audio-native processing is faster and robust to transcription errors; no figure on the pages read. | Not stated on the pages read; the API reports per-turn latency metrics (time to first token, TTS time to first byte). |
| Key limits |
|
|
|
| High-severity warnings |
|
|
|
| Compliance | /v1/live/sessions is ZDR eligible with limitations (store forced false, no forking or recording download). Abuse-monitoring logs 30 days. US and EU data residency. | Not stated on the pricing or FAQ pages (a trust portal is linked). | Not re-verified this session (Deepgram markets SOC 2 and HIPAA readiness; check the trust page). |
| Self-hostable | No | No | No |
| Last checked | 2026-10-10 | 2026-10-10 | 2026-10-10 |
| Key numbers and features | |||
| Flat $/min | $0.05 | $0.05 | $0.075 |
| Audio in $/1M tok | - | - | - |
| Audio out $/1M tok | - | - | - |
| Free tier | No | Yes | Yes |
| Free credit $ | - | - | $200 |
| Native S2S | Yes | No | No |
| Tools | Yes | Yes | Yes |
| Image in | No | - | - |
| Own LLM | Yes | - | Yes |
| Voices | 12 | - | - |
| Cloning | - | Yes | - |
| Latency ms | - | - | - |
| Languages | - | - | - |
| Context tokens | 128,000 | - | - |
| Max session min | - | - | - |
| Concurrency | 50 | 5 | 45 |
| WebRTC | Yes | Yes | No |
| WebSocket | Yes | Yes | Yes |
| Phone / SIP | Yes | Yes | No |
| HIPAA | - | - | - |
| SOC 2 | - | - | - |
| EU data | Yes | - | - |
| Open weights | No | Yes | No |
| High warnings | 2 | 1 | 2 |