Grok Voice 3.0 在 Big Bench Audio 基准上的平均首音频时间（Time to First Audio）为 0.81 秒，比 GPT-Live-1（Astra，medium）（1.34 秒）和 Gemini 3.8 Live Extended Thinking（High）（1.35 秒）快约 40%，这两者是我们 Speech to Speech Index 中紧随其后的两个模型，而 Grok Voice 3.0 仅略逊于 Grok Voice 2.0 的 0.70 秒。这使其大幅领先于 GPT-Live-1（Sol，low）（1.24 秒），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。
Grok Voice 3.0's average Time to First Audio on the Big Bench Audio benchmark is 0.81 seconds, ~40% faster than GPT-Live-1 (Astra, medium) (1.34s) and Gemini 3.8 Live Extended Thinking (High) (1.35s), the next two models in our Speech to Speech Index, and just behind Grok Voice 2.0 at 0.70 seconds. That puts it well ahead of GPT-Live-1 (Sol, low) (1.24s), with Grok Voice 3.0 and Grok Voice 2.0 the only models in the top 5 of our Speech to Speech Index under 1 second.
来源：Artificial Analysis · x.com