xAI expands Grok Voice with new multilingual voices and speech‑to‑text integration

xAI has added real‑time speech‑to‑text to its terminal‑based coding agent, Grok Build. The feature, called Grok Voice, lets developers dictate code, debugging sessions and brainstorming for up to 15 minutes using a simple “/voice” command or the Ctrl + Space shortcut. It runs on the Grok STT and TTS APIs launched in April 2026, supports more than 25 languages, offers low‑latency streaming, word‑level timestamps and speaker diarisation, and is available to SuperGrok and X Premium+ subscribers at $0.10‑$0.20 per hour.

In a related update, xAI released 21 new flagship voices for Grok Voice on 6 July 2026, expanding the original five. The voices are natively multilingual, covering 25+ languages, and include purpose‑built options such as Carina for empathetic customer‑service tones and Luna for friendly conversation. Developers can control delivery with speech tags and use the Voice Agent API for sub‑second, speech‑to‑speech interactions. A no‑code Voice Agent Builder, launched 1 July 2026 at $0.05 per minute, lets users create custom voice agents in minutes, and a voice‑cloning feature works with as little as one minute of audio. The platform topped the Big Bench Audio benchmark for voice‑agent capabilities.