SpaceXSI Launches Grok Voice 3.0, Ranks No. 1 in Speech-to-Speech AI

SpaceXSI (formerly SpacexAI) has launched Grok Voice 3.0, its latest speech-to-speech artificial intelligence model, claiming the top position on Artificial Analysis’s Speech-to-Speech Index with a score of 86.9.

Announced on October 14, the model is designed to support natural, real-time voice conversations, including customer service calls, sales interactions and AI-powered assistants.

According to SpaceXSI, the system combines advanced speech reasoning, conversational responsiveness and task completion capabilities while maintaining a competitive operating cost.

The company’s published benchmark results place Grok Voice 3.0 ahead of GPT-Live-1 (Astra, Medium), which scored 83.2, and Gemini 3.8 Live (Extended Thinking, High), which scored 82.6. The previous Grok Voice 2.0 model recorded a score of 81.3.

Credit: SpaceXSI, ArtificialAnalysis.ai

SpaceXSI says Grok Voice 3.0 delivers audio responses in under one second, helping conversations feel more natural and reducing delays during interactions. The company also reported a score of 97% on the Full Duplex Bench, which evaluates how effectively voice systems manage interruptions, pauses and conversational turn-taking.

The model achieved a 98.5% score on Speech Reasoning Big Bench Audio and a 94.8% task-success rate in the published comparison. These results suggest improvements across several aspects of voice-based AI, although performance varies by benchmark.

Grok Voice 3.0 did not lead every individual evaluation (to note here). Its reported score on τ-Voice, a benchmark focused on voice-agent performance in challenging support-call scenarios, was 71.6%. GPT-Live-1 recorded a higher score of 74.5% in the same evaluation, according to the published benchmark figures.

The distinction highlights the importance of examining individual test results alongside overall rankings when comparing competing AI systems.

SpaceXSI says Grok Voice technology already handles millions of conversations daily across Tesla and SpaceX operations. The company reports that Starlink call centres use the technology to manage approximately 20,000 customer calls per day across multiple languages, with around 70% resolved without human intervention.

According to the company, the system can assist with customer verification, technical troubleshooting, account credits and replacement-equipment requests. These operational figures are company-reported claims rather than independently audited results.

The latest release is intended to strengthen the use of voice AI in business environments where speed, accuracy and the ability to resolve requests without human assistance are important.

API Availability and Pricing

Grok Voice 3.0 became available on October 14 through SpaceXSI’s Voice Agent API at a listed price of $0.08 per minute, equivalent to $4.80 per hour, unchanged from the previous version.

The company also said applications using the grok-voice-latest model alias have been switched automatically to the new release, allowing existing users to access the updated model without changing their integration.

The launch comes as technology companies compete to develop AI systems that can listen, reason and respond in real time rather than relying solely on text-based interactions.

Grok Voice 3.0’s leading score on the overall Speech-to-Speech Index strengthens SpaceXSI’s position in this rapidly developing market. However, the results also show that no single model necessarily leads every category, making task-specific performance and real-world reliability important considerations for businesses adopting voice AI.

While the benchmark results provide an encouraging indication of the new model’s capabilities, independent testing across a broader range of languages, industries and real-world customer interactions will be important in assessing its long-term performance.

News source: X.ai/news/grok-voice-3