Voice Agent API
$0.08 / minPer minute of speech-to-speech conversation ($4.80 per hour)Text input to the session is billed separately at $0.004. Same rate as Grok Voice 2.0. Official source 1 ↗ · Official source 2 ↗
Reviewed tool profile
Grok Voice 3.0
SpaceXSI’s speech-to-speech model for real-time voice agents and support calls, at $0.08 per minute.
Sources checked
- Pricing
- Usage-based API pricing: $0.08 per minute of conversation ($4.80 per hour) plus $0.004 per text input, the same price as Grok Voice 2.0. No free API allowance for voice is listed. Official source 1 ↗ · Official source 2 ↗
- Available as
- Available through the SpaceXSI Voice Agent API at the /v1/realtime WebSocket endpoint, with model ID grok-voice-3.0 or the grok-voice-latest alias. · Client-side apps connect with Ephemeral Tokens; official demo apps cover a web agent, a Twilio phone agent, a WebRTC agent and an iOS tester app. Official source 1 ↗ · Official source 2 ↗ · Official source 3 ↗ Official source 1 ↗
- Primary group
- Voice & Speech
What it does
Grok Voice 3.0 is SpaceXSI’s (formerly xAI) speech-to-speech model for real-time voice agents, released on October 14, 2026. It takes text and audio in and returns text and audio, and runs over a WebSocket realtime endpoint with turn-taking and tool use. The company positions it for customer support calls; it also powers voice in Tesla vehicles, Grok Bot’s voice mode and Starlink’s call centers. Developers call it as grok-voice-3.0, and the grok-voice-latest alias now points to it.
Official source 1 ↗ · Official source 2 ↗ · Official source 3 ↗Features
- Real-time speech-to-speech. Holds live spoken conversations over WebSockets with sub-second latency and server-side turn detection, instead of chaining separate transcription, LLM and TTS steps. Official source 1 ↗ · Official source 2 ↗
- Tool use during calls. Voice agents can call tools such as web search in the middle of a conversation. Official source 1 ↗
- Built-in and custom voices. Ships with a roster of built-in voices that can each speak every supported language (eve is the default), and accepts cloned custom voices through the same voice_id field. Official source 1 ↗
- Handles interruptions. Vendor-reported 97% on Full Duplex Bench, which scores how naturally a model handles pauses, turn-taking and interruptions. Official source 1 ↗
- Built for support calls. A single agent at Starlink verifies customers, troubleshoots hardware, issues credits and ships replacement equipment; SpaceXSI says about 70% of 20,000 daily calls end without a human. Official source 1 ↗
- Enterprise compliance. Audio is processed in real time and not stored or used for training; the Voice APIs are SOC 2 Type II audited and HIPAA-eligible with a BAA. Official source 1 ↗
Pricing & free limits
Billing details
Text input sent into a voice session is billed at $0.004 on top of the per-minute rate. Official source 1 ↗
Agents on the grok-voice-latest alias were moved to Grok Voice 3.0 automatically at the same price; Grok Voice 2.0 is now listed as deprecated. Official source 1 ↗ · Official source 2 ↗
Free access & limits
Free access and its limits were not established by the checked official sources.
Best for & limitations
Answer and resolve customer support calls, including identity checks, troubleshooting and issuing credits. Official source 1 ↗
Build phone agents on Twilio or browser voice agents over WebRTC. Official source 1 ↗
Add hands-free voice control to an in-car or in-app assistant. Official source 1 ↗
Run multilingual call-center lines with one agent. Official source 1 ↗ · Official source 2 ↗
- Benchmarks are vendor-cited. The #1 Speech-to-Speech Index score (86.9) and EVA-Bench result come from SpaceXSI’s launch post citing Artificial Analysis and ServiceNow. Official source 1 ↗
- Not the clear leader on τ-Voice. On the τ-Voice support-call benchmark it scores 71.6%, which the company calls comparable with GPT-Live-1 rather than ahead of it. Official source 1 ↗
- Single listed region. The model page lists us-east-1 as its region. Official source 1 ↗
- Developer API only. It is an API model for building voice agents, not a ready-made app; you need to write code against the realtime endpoint. Official source 1 ↗ · Official source 2 ↗
Similar tools
Tools like Grok Voice 3.0
Published tools that share its categories.
VoxCPMOpen-source multilingual text-to-speech model from OpenBMB with voice design, voice cloning and 48kHz output.
Visit 
SiteTrellisBring-your-own-key AI chat agent for websites, with provider fallbacks, custom API tools, lead capture and live human takeover.
Visit 
OnepinAI voice production agent that routes scripts across 30+ TTS models, validates every line and auto-fixes misreads.
Visit 
Eleven v4ElevenLabs’ most expressive text-to-speech model, with a low-latency Turbo variant for voice agents.
Visit 
Claude Haiku 5.5Anthropic’s fastest, lowest-cost Claude model for high-volume tasks and coding subagents.
Visit 
MusicfyAI voice song generator for making covers in other voices, training a clone of your own voice, and generating music from text prompts.
Visit