xAI officially announces the launch of Grok Voice 3.0, claiming it to be the world's most intelligent voice-to-voice model. The model sets a new record in handling customer support calls with a latency of under one second and costs less than other leading models. Grok Voice supports millions of conversations daily and is available in the API at the same price as its predecessor, Grok Voice 2.0.
Grok Voice 3.0 achieves highest accuracy rating
According to independent testing by Artificial Analysis, Grok Voice 3.0 ranks first in the voice-to-voice index with a score of 86.9. This index combines voice reasoning, τ-Voice, as well as preferences and task success rates from blind tests. Grok Voice 3.0 scored 97% in the Full Duplex Bench, which measures the naturalness of its handling of pauses, conversations, and interruptions.
“Grok Voice 3.0 has set a new Pareto frontier on EVA-Bench, which is ServiceNow's open-source benchmark for evaluating voice agents on accuracy and customer experience.”
xAI
Grok Voice 3.0 widely used in Tesla and SpaceX
Grok Voice 3.0 processes millions of conversations daily in Tesla vehicles. Drivers can make simple requests using Connectors, such as checking their calendar or replying to emails, and can also use Grok Bot for larger tasks like ordering coffee or booking dinner, all of which can be done completely hands-free. The Starlink call center uses Grok Voice to respond to 20,000 customer calls each day, with approximately 70% of inquiries resolved without human intervention.
Pricing and availability of Grok Voice 3.0
Grok Voice 3.0 is now available in the Voice Agent API at a price of $0.08 per minute (approximately 0.62 HKD), the same as Grok Voice 2.0. Agents using `grok-voice-latest` have been automatically switched over without any changes required.
Source: xAI official announcement