SpaceXAI releases Grok Voice 3.0 speech-to-speech model: #1 on Artificial Analysis S2S Index (86.9), under 1s latency, same $0.08/min API price

Why it matters

【Key points】

SpaceXAI (formerly xAI) released Grok Voice 3.0, a speech-to-speech model, on Oct 14. It is live in the Voice Agent API as grok-voice-3.0 at the same price as 2.0: $0.08/min ($4.80/hr). Apps on the grok-voice-latest alias were switched automatically.

Citing Artificial Analysis tests dated Oct 13, it ranks #1 on the Speech-to-Speech Index at 86.9 (Grok Voice 2.0 81.3, GPT-Live-1 Astra Medium 83.2, Gemini 3.8 Live 82.6), with Big Bench Audio 98.5%, Speech Agent Arena Elo 1070, task success 94.8% and Full Duplex Bench 97.0%. On the τ-Voice support-call benchmark it scores 71.6%, below GPT-Live-1's 74.5%, but with about 40% lower latency (under 1s). Cost per task is about $0.025 vs $0.030 for GPT-Live-1 and $0.055 for GPT-Realtime-2.1 (High). It tops ServiceNow's EVA-Bench at 74% accuracy.

Deployment (company-reported): millions of daily conversations in Tesla vehicles, Grok Bot voice mode, and ~20,000 Starlink support calls a day, ~70% resolved without a human.

【Verifiable】

Announcement: https://x.ai/news/grok-voice-3

xAI models list: https://docs.x.ai/developers/models

Artificial Analysis speech-to-speech: https://artificialanalysis.ai/speech-to-speech

【Our read】

1. Voice models now compete on finishing the call, not sounding human: the headline metrics are task success, τ-Voice and cost per task, aimed at customer support.

2. #1 overall, not on every axis: it still trails GPT-Live-1 on τ-Voice by ~3 points; its edge is latency and price. Benchmarks come from Artificial Analysis and ServiceNow; Tesla/Starlink usage figures are self-reported.

3. A same-price upgrade with automatic alias switching means frictionless migration and direct price pressure on OpenAI's realtime voice models.