IT之家 · 智能时代 · 10/15/2026, 07:35:09
SpaceXAI Launches Grok Voice 3.0: Sub-Second Latency and Cost-Efficient Performance in Customer Service Benchmarks
SpaceXAI has released Grok Voice 3.0, claiming it as the world's most intelligent speech-to-speech model with a top score of 86.9 on the Artificial Analysis leaderboard. The model achieves sub-second latency and outperforms competitors in enterprise benchmarks like τ-Voice and EVA while maintaining lower operational costs. It is already deployed for Tesla vehicle interactions and Starlink customer support, with API pricing remaining unchanged at $0.08 per minute.
SOURCE COVERAGEFull Original Report
On October 15, SpaceXAI released Grok Voice 3.0, which the company claims is the world's most intelligent speech-to-speech model. The model set a new record in customer service call handling tasks, featuring industry-leading latency of under one second and lower operating costs compared to other frontier models. Grok Voice supports millions of daily conversations across Tesla and SpaceX operations and is now available via API, with pricing remaining consistent with the previous generation.
In independent testing by Artificial Analysis, Grok Voice 3.0 ranked first on the comprehensive speech-to-speech leaderboard with a score of 86.9. This leaderboard evaluates speech reasoning capabilities, τ-Voice metrics, as well as preference and task completion rates from blind call comparisons. Grok Voice 3.0 achieved a 97% score on full-duplex benchmarks, a metric that measures how naturally the model handles pauses, turn-taking, and interruptions.
τ-Voice assesses an agent's ability to handle various accents, background noise, and mid-conversation interruptions in customer service scenarios. Grok Voice 3.0 scored 71.6%, performing comparably to GPT-Live-1 (Astra) but with 40% lower latency, establishing a new optimal balance between agent intelligence and response speed.
The company states that its operating costs are also lower than those of comparable models. Artificial Analysis tracked the cost per completed task across large audio benchmarks, finding that Grok Voice 3.0 is cheaper than both GPT-Live-1 (Astra, medium version) and GPT-Realtime-2.1.
In ServiceNow's EVA benchmark for enterprise voice agents, Grok Voice 3.0 achieved the highest accuracy score of 74%, delivering the best combined performance across accuracy and customer experience metrics.
According to official reports, Grok Voice 3.0 processes millions of conversations daily within Tesla vehicles. Drivers can use Connectors to execute simple commands such as checking schedules or replying to emails, while leveraging Grok Bot for complex tasks like ordering coffee or booking dinner—all hands-free. Grok Voice 3.0 also powers the voice mode for Grok Bot.
Starlink's customer support center relies on Grok Voice to handle 20,000 incoming calls daily in multiple languages. Approximately 70% of these calls are resolved without human intervention, and one-quarter of sales calls result in orders before the customer hangs up. A single agent built on Grok Voice 3.0 can perform customer identity verification, hardware troubleshooting, issue refunds, and arrange device replacements.
Grok Voice 3.0 is now live on the voice agent API, priced at 4.80 per hour (approximately ¥32.2 CNY), matching the pricing of Grok Voice 2.0. Agents using the grok-voice-latest version have been automatically upgraded and require no changes.