We introduce Full-Duplex-Bench-v3 (FDB-v3), a benchmark for evaluating spoken language models under naturalistic speech conditions and multi-step tool use. Unlike prior work, our dataset consists entirely of real human audio annotated for five disfluency categories, paired with scenarios requiring chained API calls across four task domains. We evaluate six model configurations -- GPT-Realtime, Gemini Live 2.5, Gemini Live 3.1, Grok, Ultravox v0.7, and a traditional Cascaded pipeline (Whisper$\rightarrow$GPT-4o$\rightarrow$TTS) -- across accuracy, latency, and turn-taking dimensions. GPT-Realtime leads on Pass@1 (0.600) and interruption avoidance (13.5\%); Gemini Live 3.1 achieves the fastest latency (4.25~s) but the lowest turn-take rate (78.0\%); and the Cascaded baseline, despite a perfect turn-take rate, incurs the highest latency (10.12~s). Across all systems, self-correction handling and multi-step reasoning under hard scenarios remain the most consistent failure modes.

翻译：我们提出Full-Duplex-Bench-v3（FDB-v3），一个在自然语音条件下评估口语模型多步骤工具使用能力的基准测试。与现有工作不同，本数据集完全采用真实人类音频，标注了五种非流畅性类别，并配需跨越四个任务域进行链式API调用的场景。我们从准确率、延迟和话轮切换三个维度，评估了六种模型配置——GPT-Realtime、Gemini Live 2.5、Gemini Live 3.1、Grok、Ultravox v0.7及传统级联流水线（Whisper→GPT-4o→TTS）。GPT-Realtime在首次准确率（0.600）和打断规避（13.5%）上领先；Gemini Live 3.1取得最快延迟（4.25秒），但话轮切换率最低（78.0%）；级联基线虽拥有完美话轮切换率，却产生了最高延迟（10.12秒）。在所有系统中，自纠正处理和困难场景下的多步推理仍是最常见的失败模式。