SpaceXSI launched Grok Voice 3.0 on October 14. The speech-to-speech model is designed for voice agents that manage customer support and sales calls. It is available through the Voice Agent API for $0.08 per minute, or $4.80 per hour—the same rate listed for Grok Voice 2.0.
The company was previously known as xAI, according to SiliconANGLE. Grok Voice 3.0 enters a call-center market where employment has already been declining.
Grok Voice 3.0 Leads Artificial Analysis’ Speech Index
Artificial Analysis gave Grok Voice 3.0 a score of 86.9 on its Speech-to-Speech Index, placing it first in the ranking.
The result exceeds the reported scores for:
- GPT-Live-1 (Astra, medium): 83.2
- Google Gemini 3.8 Live: 82.6
- Grok Voice 2.0: 81.3
Grok Voice 3.0 improved by 5.6 points over Grok Voice 2.0 on the index. It also ranked first for task success, with a reported score of 94.8%, and received 98.5% on Big Bench Audio reasoning.
In head-to-head preference voting, the model ranked third with an Elo score of 1,070.
Speech Accuracy, Latency, and Benchmark Results
The model performed well across several tests, though it did not lead in every measurement.
According to the reported results, Grok Voice 3.0 achieved the highest accuracy score on ServiceNow’s EVA-Bench for enterprise voice agents, reaching 74%. The model produced its first audio in an average of 0.81 seconds. That was faster than the competing models cited in the comparison, although it was slower than Grok Voice 2.0, which averaged 0.70 seconds.
On Tau-Voice, a benchmark based on support calls involving strong accents and background noise, OpenAI’s model held a narrow advantage:
|
Model
|
Tau-Voice Score
|
|
OpenAI model
|
74.5%
|
|
Grok Voice 3.0
|
71.6%
|
So, while Grok Voice 3.0 led the overall Speech-to-Speech Index, OpenAI’s model scored higher on this particular support-call benchmark.
Grok Voice 3.0 Pricing and Estimated Task Costs
SpaceXSI lists Grok Voice 3.0 at $0.08 per minute through its Voice Agent API. That works out to $4.80 per hour and matches the stated price of Grok Voice 2.0.
Artificial Analysis estimated the cost of a single task at roughly $0.025. The reported comparisons were:
|
Model
|
Estimated Cost per Task
|
|
Grok Voice 3.0
|
$0.025
|
|
GPT-Live-1
|
$0.030
|
|
GPT-Realtime-2.1
|
$0.055
|
Hourly pricing comparisons reported elsewhere placed Gemini 3.8 Live at $3.50 per hour and GPT-Live-1 Astra at $5.83 per hour. Grok Voice 3.0 sits between those prices at $4.80 per hour.
Use in Starlink Call Centers and Tesla Vehicles
SpaceXSI says Starlink call centers use Grok Voice for around 20,000 calls each day. According to the company, about 70% of those calls are resolved without a human.
The company also says that one in four sales calls results in a purchase. A single agent can verify customers, troubleshoot hardware, issue credits, and arrange replacement-equipment shipments.
Grok Voice is also said to handle millions of conversations per day in Tesla vehicles.
Automatic Upgrade for Existing Voice Agent API Users
Developers using the grok-voice-latest alias were automatically moved to Grok Voice 3.0. That means voice agents already configured with that alias now use the new model without a separate change from the developer.
Grok Voice 3.0 Arrives as Call-Center Employment Declines
The launch comes during a period of falling call-center employment. An October 6 report cited by TheStreet found that global call-center employment was 5.3% below its December 2023 peak and had declined year over year for eight consecutive quarters.
TheStreet also reported that Salesforce reduced its customer-service workforce from 9,000 to roughly 5,000 with help from AI.