Elon Musk’s SpaceXSI has launched the new Grok Voice 3.0 speech-to-speech model designed for voice agents in customer service and sales centers. Upon release, the model immediately took first place on the Artificial Analysis index, outperforming models from OpenAI and Google.
Grok Voice 3.0 launched for voice agents at the same pricing as the previous generation
Top ranking in benchmark metrics and a comparison of speed and cost with competitors
Widespread field applications in Starlink call centers and Tesla vehicles
Ongoing contraction in the global call center industry due to automation
Frequently asked questions
Grok Voice 3.0 launched for voice agents at the same pricing as the previous generation
On October 14, 2026, billionaire Elon Musk’s artificial intelligence company launched its new voice model, “Grok Voice 3.0.” This advanced model, which operates directly via speech-to-speech technology, was designed to power smart voice agents handling customer support and sales calls. Upon its debut, it successfully captured the number one spot on the “Artificial Analysis” index for conversational voice models.
The company’s official launch post stated that the new model is now available to developers via the voice agent API at a price of $0.08 per minute, or $4.80 per hour—the exact same pricing used for the previous version, “Grok Voice 2.0.”
SiliconANGLE reported that the company now operates under the name “SpaceXSI,” having previously been known as “xAI,” noting that its marketing pitch targets the global call center sector, which is already experiencing a continuous contraction in workforce numbers.
Top ranking in benchmark metrics and a comparison of speed and cost with competitors
Technical evaluation firm Artificial Analysis awarded “Grok Voice 3.0” an overall score of 86.9 points on its benchmark index, outperforming the mid-tier “GPT Live 1 Astra” model at 83.2 points and Google’s “Gemini 3.8 Live” model at 82.6 points. It also achieved a 5.6-point jump compared to the previous version, “Grok Voice 2.0,” which scored 81.3 points. The model ranked first in task success rate at 94.8 percent, scored 98.5 percent on the Big Bench Audio test for audio reasoning, and placed third in direct preference voting with an Elo rating of 1,070 points.
However, the model did not top every test; OpenAI retained a slight lead in the “Tau Voice” benchmark, which is built from real support calls featuring heavy accents and background noise, scoring 74.5 percent compared to Grok’s 71.6 percent, according to a SiliconANGLE report. Grok Voice 3.0 produced its first audio response in an average of 0.81 seconds, a speed faster than its competitors but slightly slower than Grok Voice 2.0’s 0.70 seconds. The company also confirmed that the model achieved a top accuracy of 74 percent on ServiceNow’s “Eva Bench” for enterprise voice agents.
In terms of operational cost, Artificial Analysis estimated the cost per task for Grok at approximately $0.025, compared to $0.030 for GPT Live 1 and $0.055 for GPT Realtime 2.1, according to The Street. On an hourly pricing basis, Grok’s cost ($4.80) is higher than Gemini 3.8 Live ($3.50) but lower than GPT Live 1 Astra ($5.83).
Widespread field applications in Starlink call centers and Tesla vehicles
According to data announced by the company, Starlink satellite internet call centers use the Grok Voice model to handle around 20,000 calls daily, successfully resolving nearly 70 percent of those calls entirely without human intervention.
The company also stated that one out of every four sales calls ends in an actual purchase, according to The Street report. A single voice agent can verify customer identities, diagnose and repair hardware malfunctions, issue corrective financial credits, and ship replacement equipment to customers.
The company added that the model also manages millions of voice conversations daily inside Tesla electric vehicles, as noted by Unite.AI, while developers using the codename for the latest voice release were upgraded to the new model automatically.
Ongoing contraction in the global call center industry due to automation
The rollout of this advanced voice model arrives at a time when total global employment in the call center sector has dropped by 5.3 percent from its peak recorded in December 2023.
Employment volume in this sector has registered a year-over-year decline for eight consecutive quarters, according to a report published on October 6 and cited by The Street.
The same article pointed to a prominent example of this corporate shift, explaining that software giant Salesforce scaled back its customer service workforce from 9,000 employees to approximately 5,000 employees with the help of artificial intelligence tools.
Frequently asked questions
Question: What rank did Grok Voice 3.0 achieve in global indices?
Answer: The model took first place on the Artificial Analysis conversational voice index with a score of 86.9, first place in task success at 94.8 percent, and 74 percent accuracy on the Eva Bench standard.
Question: How much does it cost developers and businesses to use Grok Voice 3.0?
Answer: The usage price is $0.08 per minute or $4.80 per hour, with the cost per individual task estimated at approximately $0.025.
Question: How is the Grok Voice model used practically at Starlink and Tesla?
Answer: In Starlink call centers, it handles around 20,000 calls daily and resolves 70 percent of them without human intervention, and it conducts millions of daily conversations inside Tesla vehicles.
Question: How has artificial intelligence affected employment volume in the call center sector?
Answer: Global call center employment has declined by 5.3 percent from its December 2023 peak across eight consecutive quarters, and Salesforce reduced its support staff from 9,000 to about 5,000 employees.
