Grok Voice 3.0 leads speech benchmark; OpenAI updates GPT-Live to 1.1

xAI's Grok Voice 3.0 scored 86.9 on Artificial Analysis's Speech-to-Speech Index, topping OpenAI's GPT-Live-1 (83.2) and Google's voice models on speed, task success, and cost. On the τ-Voice benchmark simulating customer service calls, Grok Voice 3.0 achieved a 71.6% task success rate, up from 56.5% with Grok Voice 2.0. It responds with roughly 0.8-second latency versus GPT-Live-1's 1.34 seconds, and costs $0.025 per task compared to $0.030 for GPT-Live-1 and $0.055 for GPT-Realtime-2.1. The model is deployed inside Tesla vehicles handling hands-free tasks including calendar checks, email replies, and booking services. OpenAI separately released GPT-Live-1.1, rolling out to all ChatGPT, Codex, and Dots users, with improvements to voice quality, better accents, reduced filler phrases, and new voices in French, German, Hindi, Mexican Spanish, Korean, and Japanese; API access is described as coming soon.