Phonic 1（High）以 79.9% 登顶 Tau Voice——我们评估智能体表现的基准，领先第二名 GPT-Live-1（Astra，medium，74.5%）5.4 个百分点，其后为 Grok Voice 3.0（71.6%）、Gemini 3.8 Live Extended Thinking（High，68.6%）和 GPT-Live-1（Sol，low，59.3%）。与 Arena 中反映与真人较短实时对话的任务成功率不同，Tau Voice 测试的是长时客服任务中严格遵循策略与多步工具使用。
Phonic 1 (High) takes the #1 spot on Tau Voice, our benchmark for agentic performance, at 79.9%, a clear lead of 5.4 points over the next model, GPT-Live-1 (Astra, medium) at 74.5%, Grok Voice 3.0 at 71.6%, Gemini 3.8 Live Extended Thinking (High) at 68.6% and GPT-Live-1 (Sol, low) at 59.3%. Unlike Task Success Rate in our Arena, which reflects shorter live conversations with real people, Tau Voice tests strict policy-following and multi-step tool use in long customer-service tasks.
来源：Artificial Analysis · x.com