xAI's New Voice Model Dominates Benchmarks, Outpacing Gemini and GPT Realtime

xAI has released grok-voice-think-fast-1.0, a full-duplex voice AI model that tops the τ-voice Bench with a score of 67.3%, significantly outperforming competitors like Gemini 3.1 Flash Live (43.8%) and GPT Realtime 1.5 (35.3%). The model is designed for complex, multi-step voice interactions in noisy environments, handling interruptions, accents, and natural turn-taking. It is already powering Starlink's live phone operations. In vertical benchmarks, the model leads in retail (62.3%), airline (66%), and telecom (73.7%) scenarios, showcasing its robustness across customer support, sales, and enterprise applications.