SpaceXSI's Grok Voice 3.0 Tops Industry Benchmarks in Speech-to-Speech Showdown
Elon Musk's SpaceXSI has unveiled Grok Voice 3.0, a speech-to-speech model that outperforms competitors from OpenAI and Google across multiple independent testing frameworks.

SpaceXSI, the artificial intelligence division of Elon Musk's enterprise formerly known as xAI, unveiled Grok Voice 3.0 today. The updated speech-to-speech model powers the company's Voice Agent API and targets enterprises managing customer interactions and sales conversations through voice technology.
Testing by independent firm Artificial Analysis placed Grok Voice 3.0 atop its Speech-to-Speech Index with a score of 86.9. The ranking methodology incorporates reasoning assessments, task completion rates and listener preferences gathered through blind comparative evaluations. OpenAI Group PBC's GPT-Live-1, paired with the GPT-6 Astra reasoning model, achieved 83.2 on the same measure. Google LLC's Gemini 3.8 Live scored 82.6.
SpaceXSI's model demonstrated particular strength in handling natural conversation dynamics. On Full Duplex Bench, which evaluates how well systems manage pauses and overlapping speech, Grok Voice 3.0 reached 97%. The Tau-Voice benchmark, designed around support scenarios featuring heavy accents and ambient noise, showed OpenAI's offering slightly ahead at 74.5% versus 71.6%, though SpaceXSI reported its model delivered initial audio roughly 40% faster in that test.
Cost efficiency also favored the SpaceXSI offering. Running a task through Artificial Analysis' Big Bench Audio suite cost approximately 2.5 cents on Grok Voice 3.0, compared with 3 cents for GPT-Live-1 and 5.5 cents for OpenAI's GPT-Realtime-2.1.
Beyond Artificial Analysis' assessments, Grok Voice 3.0 led on ServiceNow Inc.'s EVA-Bench, an open-source evaluation covering more than 200 enterprise use cases, where it achieved 74% accuracy. The model showed no peers when combining accuracy with customer experience metrics. Nitin Aggarwal, senior director of product management at ServiceNow, stated that the model "pairs top accuracy with a natural, responsive conversation."
Real-world deployment
Tesla Inc. vehicles already generate millions of daily Grok Voice interactions. Drivers use the system for calendar checks and email management, while more complex tasks like restaurant reservations route through Grok Bot, which now runs on the new model.
At Starlink, SpaceX's satellite internet operation, Grok Voice handles approximately 20,000 customer calls daily across multiple languages. SpaceXSI reported that roughly 70% of these interactions resolve without human intervention, and one in four sales calls result in a purchase completion before the call ends.
Availability and pricing
Grok Voice 3.0 launched today through the Voice Agent API at eight cents per minute, equivalent to $4.80 hourly—matching the pricing of Grok Voice 2.0. Developers using the grok-voice-latest model alias have been automatically transitioned to the new version without requiring any modifications to their implementations.
SpaceX acquired xAI in February. The unit was rebranded to SpaceXAI in July and adopted the SpaceXSI designation earlier this month.


