Voice, Speech & Realtime AI

xAI ships Grok Voice Transcribe 2.0 and Grok 4.7, keeping pace in speech and frontier models

Source: AI Weekly · Sep 22, 2026

xAI has shipped two releases in quick succession: Grok Voice Transcribe 2.0 on September 18, reported to roughly double transcription accuracy at the same price, and Grok 4.7, its latest frontier model, on September 21. Both are tightly integrated with X, and together they show xAI keeping pace on two fronts at once — speech and general-purpose reasoning.

The transcription upgrade is the quietly consequential one. Cheaper, more accurate speech-to-text keeps lowering the cost of converting spoken conversation into structured, machine-usable text at scale. Every incremental gain there makes it easier to capture and process the spoken word, which is where a great deal of the most valuable and most sensitive human data lives.

That cuts both ways. Better, cheaper transcription expands what can be done with voice data enormously — but it also means far more recorded human conversation becomes trivially easy to ingest, index and reuse. The technical barrier to using spoken data keeps falling; the barrier that does not fall on its own is the one that matters most — whether that conversation was captured with consent and stripped of the personal detail that makes it unsafe to reuse.

Key Points

• Grok Voice Transcribe 2.0 shipped 2026-09-18 — reported roughly 2x accuracy at the same price
• Grok 4.7 released 2026-09-21
• xAI is pushing on both speech (transcription) and frontier models, tightly integrated with X
• Cheaper, more accurate transcription lowers the cost of turning spoken conversation into usable data