Grok Voice 3.0 Tops Speech Index with 86.9 Score
SpaceXSI’s new voice model ranks first on the Artificial Analysis index, outperforming OpenAI and Google in speed and cost efficiency for customer support tasks.
Key points
- Grok Voice 3.0 scored 86.9 on the Speech-to-Speech Index, beating OpenAI and Google models.
- The model resolves 70% of Starlink calls without human intervention.
- Task costs are lower at 2.5 cents compared to competitors' 3 to 5.5 cents.
SpaceXSI released Grok Voice 3.0, a speech-to-speech model that independent testers rank as the best in its class for handling customer support and sales calls.
The risk
Independent testing firm Artificial Analysis rated Grok Voice 3.0 at 86.9 on its Speech-to-Speech Index. This score surpasses OpenAI Group PBC’s GPT-Live-1, which scored 83.2, and Google LLC’s Gemini 3.8 Live, which landed at 82.6. The rankings rely on reasoning tests, task success rates, and listener votes from blind head-to-head calls. On the Full Duplex Bench, which measures how naturally a model handles pauses and interruptions, the new Grok model achieved 97%.
Who is exposed
Tesla Inc. vehicles already use Grok Voice for millions of daily conversations, including calendar checks and email responses. At Starlink, the satellite internet business, the model answers about 20,000 customer calls daily in multiple languages. SpaceXSI states that roughly 70% of these calls are resolved without human involvement. Additionally, one in four sales calls on the platform results in a purchase before the caller disconnects.
Cost and compliance
Grok Voice 3.0 costs about 2.5 cents to complete a task in Artificial Analysis’ Big Bench Audio run. This is lower than the 3 cents charged by GPT-Live-1 and the 5.5 cents for OpenAI’s GPT-Realtime-2.1. On ServiceNow Inc.’s open-source EVA-Bench, the model scored 74% accuracy across more than 200 enterprise scenarios. No other model combined higher accuracy with better customer experience scores, according to the source material.
What should you do?
Evaluate current voice agent vendors for latency and cost metrics before committing to long-term contracts. Monitor independent benchmarking reports to verify performance claims against your specific industry requirements.
The background in plain language
Teams usually assess voice agents by their ability to understand context and respond naturally. Check whether the system handles interruptions well, as this affects user trust. Verify that the model can resolve common issues without human help to reduce support costs. Ensure the vendor provides clear data on accuracy and latency, as these factors directly impact customer satisfaction and operational efficiency in high-volume environments.
What to do and how to stay safe: Grok Voice 3.0
- Review current vendor contracts for performance guarantees related to call resolution rates.
- Test new voice models with your specific customer support scripts before deployment.
- Monitor latency metrics to ensure responses feel natural to callers.
- Verify that the model handles background noise and accents effectively in your region.
Step-by-step guide: AI Governance Explained: Building Rules for Autonomous Systems
General security guidance from the Vector Tribune newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
How does Grok Voice 3.0 compare to OpenAI and Google models?
It scored higher than both on the Artificial Analysis Speech-to-Speech Index, with 86.9 versus 83.2 and 82.6.
What is the cost of using Grok Voice 3.0?
A task in the Big Bench Audio run cost about 2.5 cents, compared to 3 cents for GPT-Live-1.
How many calls does Grok Voice handle at Starlink?
It answers about 20,000 customer calls a day, resolving 70% without human intervention.
