SpaceXSI, formerly known as xAI, has launched Grok Voice 3.0, an updated speech-to-speech model for its Voice Agent API. According to Artificial Analysis, the model achieved a score of 86.9 on the Speech-to-Speech Index, surpassing OpenAI's GPT-Live-1 at 83.2 and Google's Gemini 3.8 Live at 82.6.
The model recorded a 97% score on the Full Duplex Bench, which measures natural handling of pauses and interruptions. In tests involving background noise and accents, Grok Voice 3.0 reached first audio 40% faster than the OpenAI model. Regarding costs, a task in the Big Bench Audio run costs $0.025 with Grok Voice 3.0, compared to $0.03 for GPT-Live-1 and $0.055 for GPT-Realtime-2.1.
ServiceNow's EVA-Bench, which evaluated over 200 enterprise scenarios, ranked Grok Voice 3.0 first for accuracy at 74%. The model is currently utilized in Tesla vehicles for daily tasks and by Starlink, where it handles approximately 20,000 customer calls daily. SpaceXSI reports that 70% of these calls are resolved without human intervention, and 25% of sales calls result in a purchase.
SpaceX acquired xAI in February, rebranded the unit to SpaceXAI in July, and adopted the name SpaceXSI this month. Grok Voice 3.0 is available via the Voice Agent API at a rate of $0.08 per minute or $4.80 per hour. Developers using the grok-voice-latest alias have been migrated to the new version automatically.
