xAI has released Grok Voice 3.0, a new speech-to-speech artificial intelligence model. The company calls it the most intelligent voice model available.
The model is already in use across Tesla and SpaceX, where it powers millions of conversations each day. It is also available now to developers through an API.
Grok Voice 3.0 Leads Independent Benchmarks
In testing by research firm Artificial Analysis, Grok Voice 3.0 ranked first on the Speech-to-Speech Index. It scored 86.9, ahead of OpenAI’s GPT-Live-1 at 83.2 and Google’s Gemini 3.8 Live at 82.6.
The index combines speech reasoning, customer support call tests, and results from blind head-to-head calls. The previous version, Grok Voice 2.0, scored 81.3.
On the Big Bench Audio reasoning test, the new model scored 98.5%. Its task success rate was 94.8%, compared with 87.4% for GPT-Live-1 and 89.1% for Gemini 3.8 Live.
The model also scored 97% on Full Duplex Bench. This test measures how naturally a voice model handles pauses, turn-taking, and interruptions.
On the τ-Voice test, which looks at support calls with different accents and background noise, Grok Voice 3.0 scored 71.6%. GPT-Live-1 scored slightly higher at 74.5%, but xAI said its model responds with 40% lower latency.
xAI said the model responds in under one second. The company said that speed leads its class.
Cost was also measured. Artificial Analysis found Grok Voice 3.0 costs about $0.025 per task, compared with $0.030 for GPT-Live-1 and $0.055 for GPT-Realtime-2.1.
ServiceNow also tested the model on EVA-Bench, its benchmark for business voice agents. Grok Voice 3.0 posted the highest accuracy score at 74%.
“It pairs top accuracy with a natural, responsive conversation,” said Nitin Aggarwal, Senior Director of Product Management at ServiceNow. The benchmark covers more than 200 realistic business scenarios.
How Tesla and Starlink Use Grok Voice
Grok Voice 3.0 handles millions of conversations every day inside Tesla vehicles. Drivers can use it to check their calendar or answer an email without using their hands.
For bigger tasks, drivers can use Grok Bot to order coffee or book a dinner reservation. Grok Voice 3.0 also powers Grok Bot’s voice mode.
Starlink, the satellite internet service run by SpaceX, uses Grok Voice in its call centers. The system answers about 20,000 customer calls each day in several languages.
Roughly 70% of those calls are resolved without a human worker. One in four sales calls ends with a purchase before the caller hangs up.
A single agent built on the model can verify customers, troubleshoot hardware, issue credits, and ship replacement equipment.
Grok Voice 3.0 is available now through xAI’s Voice Agent API. It costs $0.08 per minute, or $4.80 per hour, the same price as Grok Voice 2.0.
Developers using the “grok-voice-latest” setting have already been moved to the new model automatically. No changes are needed on their end.
