xAI launched Grok Voice 3.0 on October 14, 2026, offering developers an upgraded speech-to-speech model through its Voice Agent API at $0.08 per minute.
The company says the model combines sub-second response latency, improved conversational performance, and lower operating costs.
It targets applications such as customer service, hands-free vehicle assistance, and multilingual voice automation.
Benchmark Performance
According to results cited by xAI from Artificial Analysis, Grok Voice 3.0 scored 86.9 on the Speech-to-Speech Index, ranking ahead of the evaluated competitors.
The comparison placed GPT-Live-1 (Astra, Medium) at 83.2, Gemini 3.8 Live (Extended Thinking, High) at 82.6, and Grok Voice 2.0 at 81.3.
The model recorded 98.5% on Big Bench Audio speech reasoning, a 1,070 Elo rating in Speech Agent Arena, a 94.8% task success rate, and 97.0% on Full Duplex Bench.
Its predecessor scored 97.2%, 1,011 Elo, 94.6%, and 95.1%, respectively, on those measures, according to the published comparison.
Artificial Analysis also reported a 71.6% result for Grok Voice 3.0 on tau-Voice, which evaluates performance in simulated customer support conversations involving different accents, background noise, and interruptions.
xAI said the model delivered comparable performance to GPT-Live-1 (Astra) with 40% lower latency.
On the Big Bench Audio evaluation, the reported average cost per task was $0.025, compared with $0.030 for GPT-Live-1 (Astra, Medium) and $0.055 for GPT-Realtime-2.1 (High).
Enterprise testing provided another performance measure.
xAI said Grok Voice 3.0 achieved 74% accuracy on ServiceNow's EVA-Bench and ranked highest when accuracy and customer experience were assessed together across realistic enterprise scenarios.
Nitin Aggarwal, Senior Director of Product Management for Agentic Evals and RAI at ServiceNow, described the model's performance in a statement accompanying the announcement.
"Grok Voice 3.0 sets a new Pareto frontier on EVA-Bench, ServiceNow’s open-source benchmark scoring voice agents on accuracy and experience across 200+ realistic enterprise scenarios.
It pairs top accuracy with a natural, responsive conversation, setting a new standard for what enterprises can expect from AI on the front line of customer service."
Aggarwal said.
EVA-Bench evaluates voice agents across customer service, IT service management, and human resources.
Its framework assesses whether an agent completes a task correctly and how effectively it manages the spoken interaction, according to ServiceNow AI Research.
xAI also cited existing deployments to demonstrate the system's operational use.
The company said Grok Voice handles millions of daily conversations in Tesla vehicles, where drivers can use voice commands to check calendars, respond to emails, order coffee, and arrange dinner without using their hands.
At Starlink call centers, the company said its voice technology handles approximately 20,000 customer calls daily across multiple languages.
About 70% of those calls are resolved without human intervention, while one in four sales calls reportedly results in a purchase before the conversation ends.
These deployment figures were provided by xAI and were not independently verified in the supplied reporting.
The release also includes a transition for existing developers.
According to xAI, applications using the grok-voice-latest model identifier have been migrated automatically, without requiring code changes.
The API retains the previous pricing of $0.08 per minute, equivalent to $4.80 per hour, while the October 14 release notes identify Grok Voice 3.0 as the current flagship speech-to-speech model.
