Grok Voice 3.0
xAI / Grok · 2026-10-14 · 类别未确认
官方发布来源 · 按需求选型 · 查看覆盖缺口 · 归档与转录记录
Grok Voice 3.0
xAI 发布 Grok Voice 3.0，官方定位为「世界上最智能的 speech-to-speech 模型」：在 Artificial Analysis 独立评测中登上语音综合指数第一，延迟低于一秒，成本低于其他前沿模型。已收录 7 条评测记录，集中在语音智能体能力与客户体验，亮点为 AA Speech-to-Speech Index 86.9 与 Big Bench Audio 98.5%。
- 输入模态
- 音频 / 文本
- 上下文
- 官方资料未说明
- 参数
- 官方资料未说明
- 价格（每百万 tokens）
- USD 输入 未说明 / 输出 未说明 · Voice Agent API 按分钟计费：$0.08 / 分钟（$4.80 / 小时），与前代 Grok Voice 2.0 同价
本变体的评测证据
归一化读数：不计算；样本及方差不完整，无法计算置信区间。
原文位置与完整协议
heading: The Most Accurate Voice Model · table: Speech-to-speech benchmark comparison (per Artificial Analysis) · row: Overall / AA Speech-to-Speech Index · quote_snippet: Overall AA Speech-to-Speech Index | 86.9
{
"harness": "Artificial Analysis 独立评测",
"tools": null,
"shots": null,
"reasoning_effort": null,
"temperature": null,
"top_p": null,
"token_budget": null,
"turn_limit": null,
"time_limit": null,
"run_count": null,
"aggregation": null,
"judge": null
}表格 caption 标注 per Artificial Analysis，来源行标注 Source: Artificial Analysis, October 13, 2026，故按第三方转述记。正文复述该值：'ranks #1 on the Speech-to-Speech Index, with a score of 86.9'。同表竞品列：Grok Voice 2.0 81.3 / GPT-Live-1 (Astra, Medium) 83.2 / Gemini 3.8 Live (Ext. Thinking, High) 82.6；其中 Gemini 的 82.6 与 Google 自家发布页记录的 82.6 一致，可确认两边引用同一份 AA 评测。new-benchmark
归一化读数：不计算；样本及方差不完整，无法计算置信区间。
原文位置与完整协议
heading: The Most Accurate Voice Model · table: Speech-to-speech benchmark comparison (per Artificial Analysis) · row: Speech Reasoning / Big Bench Audio · quote_snippet: Speech Reasoning Big Bench Audio | 98.5%
{
"harness": "Artificial Analysis 独立评测",
"tools": null,
"shots": null,
"reasoning_effort": null,
"temperature": null,
"top_p": null,
"token_budget": null,
"turn_limit": null,
"time_limit": null,
"run_count": null,
"aggregation": null,
"judge": null
}表格 caption 标注 per Artificial Analysis，故按第三方转述记。同表竞品列：Grok Voice 2.0 97.2% / GPT-Live-1 (Astra, Medium) 90.1% / Gemini 3.8 Live (Ext. Thinking, High) 97.7%；其中 Gemini 的 97.7% 与 Google 自家发布页记录的 97.7% 一致。同页另有一张内联 SVG 成本图，给出该基准的每任务平均成本（USD，越低越好）：Grok Voice 3.0 $0.025 / GPT-Live-1 (Astra, Medium) $0.03 / GPT-Realtime-2.1 (High) $0.055——成本属对照而非基准分数，按抽取口径不单独建行。new-benchmark
归一化读数：不计算；样本及方差不完整，无法计算置信区间。
原文位置与完整协议
heading: The Most Accurate Voice Model · table: Speech-to-speech benchmark comparison (per Artificial Analysis) · row: Human Preference / Speech Agent Arena Elo · quote_snippet: Human Preference Speech Agent Arena Elo | 1070
{
"harness": "Artificial Analysis 独立评测（真人盲测）",
"tools": null,
"shots": null,
"reasoning_effort": null,
"temperature": null,
"top_p": null,
"token_budget": null,
"turn_limit": null,
"time_limit": null,
"run_count": null,
"aggregation": null,
"judge": null
}表格 caption 标注 per Artificial Analysis，故按第三方转述记。同表竞品列：Grok Voice 2.0 1011 / GPT-Live-1 (Astra, Medium) 1048 / Gemini 3.8 Live (Ext. Thinking, High) 990。new-benchmark
归一化读数：不计算；样本及方差不完整，无法计算置信区间。
原文位置与完整协议
heading: The Most Accurate Voice Model · table: Speech-to-speech benchmark comparison (per Artificial Analysis) · row: Task Completion / Task Success Rate · quote_snippet: Task Completion Task Success Rate | 94.8%
{
"harness": "Artificial Analysis 独立评测",
"tools": null,
"shots": null,
"reasoning_effort": null,
"temperature": null,
"top_p": null,
"token_budget": null,
"turn_limit": null,
"time_limit": null,
"run_count": null,
"aggregation": null,
"judge": null
}表格 caption 标注 per Artificial Analysis，故按第三方转述记。该行是 AA Speech-to-Speech Index 的任务完成分项。同表竞品列：Grok Voice 2.0 94.6% / GPT-Live-1 (Astra, Medium) 87.4% / Gemini 3.8 Live (Ext. Thinking, High) 89.1%。new-benchmark
归一化读数：不计算；样本及方差不完整，无法计算置信区间。
原文位置与完整协议
heading: The Most Accurate Voice Model · table: Speech-to-speech benchmark comparison (per Artificial Analysis) · row: Conversational Dynamics / Full Duplex Bench · quote_snippet: Conversational Dynamics Full Duplex Bench | 97.0%
{
"harness": "Artificial Analysis 独立评测",
"tools": null,
"shots": null,
"reasoning_effort": null,
"temperature": null,
"top_p": null,
"token_budget": null,
"turn_limit": null,
"time_limit": null,
"run_count": null,
"aggregation": null,
"judge": null
}表格 caption 标注 per Artificial Analysis，故按第三方转述记。正文复述该值：'Grok Voice 3.0 also scores 97% on Full Duplex Bench, a measure of how naturally it handles pauses, turn-taking, and interruptions'。同表竞品列：Grok Voice 2.0 95.1% / GPT-Live-1 (Astra, Medium) 94.9% / Gemini 3.8 Live (Ext. Thinking, High) 91.9%。new-benchmark
归一化读数：不计算；样本及方差不完整，无法计算置信区间。
原文位置与完整协议
heading: The Most Accurate Voice Model · row: τ-Voice · quote_snippet: Grok Voice 3.0 scores 71.6%, comparable with GPT-Live-1 (Astra), but at 40% lower latency
{
"harness": "Artificial Analysis 独立评测",
"tools": null,
"shots": null,
"reasoning_effort": null,
"temperature": null,
"top_p": null,
"token_budget": null,
"turn_limit": null,
"time_limit": null,
"run_count": null,
"aggregation": null,
"judge": null
}数值出现在正文散文中（DOM 明文），同页配图「Agentic Performance vs. Speed τ-Voice」标注 Source: Artificial Analysis，故按第三方转述记。官方称与 GPT-Live-1 (Astra) 相当，但延迟低 40%。本行与 Google Gemini 3.8 Live 页自报的 τ-Voice 68.6%（vendor_reported）协议不同，comparison_scope = only_same_protocol，跨厂商比较需注意口径差异。
归一化读数：不计算；样本及方差不完整，无法计算置信区间。
原文位置与完整协议
heading: The Most Accurate Voice Model · row: EVA-Bench · quote_snippet: On EVA-Bench, ServiceNow's benchmark for enterprise voice agents, Grok Voice 3.0 achieves the highest accuracy score at 74%
{
"harness": null,
"tools": null,
"shots": null,
"reasoning_effort": null,
"temperature": null,
"top_p": null,
"token_budget": null,
"turn_limit": null,
"time_limit": null,
"run_count": null,
"aggregation": null,
"judge": null
}EVA-Bench 是 ServiceNow 的开源企业语音智能体基准（覆盖 200+ 真实企业场景，同时评准确率与客户体验），数值印在 xAI 发布页正文，故按自报记；同页引用了 ServiceNow 高级产品总监 Nitin Aggarwal 的背书原文。new-benchmark