Grok Voice 3.0在Big Bench Audio基准测试中的平均Time to First Audio为0.81秒，比我们Speech to Speech Index中紧随其后的两个模型GPT-Live-1 (Astra, medium) (1.34s)和Gemini 3.8 Live Extended Thinking (High) (1.35s)快约40%，仅略低于0.70秒的Grok Voice 2.0。这使其远超GPT-Live-1 (Sol, low) (1.24s)，并且Grok Voice 3.0和Grok Voice 2.0是我们Speech to Speech Index前5名中仅有的低于1秒的模型。
