xAI has released Grok Voice 3.0 through its Voice Agent API today, charging $0.08 per minute, or $4.80 per hour, unchanged from Grok Voice 2.0. The company says the speech-to-speech model responds in less than one second and improves its performance on customer support calls. Agents configured to use grok-voice-latest have already received the upgrade automatically, without developers needing to make changes.
According to independent Artificial Analysis testing cited by xAI, Grok Voice 3.0 leads the Speech-to-Speech Index with 86.9 points, compared with its predecessor's 81.3. That index draws on speech reasoning, support-call testing and blind comparisons that assess preferences and successful task completion. On Full Duplex Bench, which tests conversational behavior including interruptions and pauses, the new model scored 97%, against 95.1% for Grok Voice 2.0.
The support-call test puts it behind one rival: Grok Voice 3.0 scored 71.6% on τ-Voice, below the 74.5% achieved by GPT-Live-1 (Astra, Medium). xAI nevertheless says its model's latency is 40% lower than GPT-Live-1 (Astra). The test subjects agents to accents, noisy surroundings and interruptions during support conversations.
Artificial Analysis also measured an average task cost of $0.025 for Grok Voice 3.0 in its Big Bench Audio testing. That compares with $0.030 for GPT-Live-1 (Astra, Medium) and $0.055 for GPT-Realtime-2.1 (High). Separately, xAI reports 74% accuracy on ServiceNow's EVA-Bench, placing Grok Voice 3.0 first for accuracy and for the combined assessment of accuracy and customer experience.
xAI says Grok Voice 3.0 already handles millions of daily conversations inside Tesla vehicles and supplies the voice mode for Grok Bot. Drivers can use Connectors to check calendars or answer email, while Grok Bot can handle requests such as dinner reservations. Starlink uses Grok Voice for 20,000 customer calls daily across several languages, with roughly 70% resolved without human involvement, according to xAI.
The release follows Grok Voice Transcribe 2.0, which xAI introduced in September with a claim of twice its predecessor's accuracy at unchanged prices. That speech-to-text product charges $0.10 per audio hour for batch processing and $0.20 per hour for streaming transcription.
