xAI's Grok Voice Transcribe 2.0 keeps price, claims better accuracy

xAI released Grok Voice Transcribe 2.0 on 18 September 2026, a speech-to-text model the company says roughly doubles the accuracy of its predecessor at unchanged pricing: $0.10 per hour of audio for batch transcription and $0.20 per hour for streaming. Diarization, timestamps and key-term biasing remain included at no extra cost.

What xAI is claiming

The company reports a word error rate of 6.8% on short phrases, down from 20.6% for version 1.0, and says the new model ranks first for accuracy among 32 streaming models on the Artificial Analysis leaderboard. Loom is named as a design partner for the release. None of these figures have been independently reproduced; they come from xAI's own announcement and from a single third-party leaderboard xAI chose to cite.

Why the flat price is the more useful signal

Speech-to-text pricing has been falling across vendors for months, so xAI matching its own previous price while claiming a large accuracy jump is a more checkable data point than the accuracy number itself: the API call signature and cost model are unchanged, which lowers the switching cost for teams already integrated with version 1.0.

What to do before flipping the default

The same /v1/stt endpoint accepts either grok-voice-transcribe-1.0 or grok-voice-transcribe-2.0, and xAI has said 2.0 will become the API default with 1.0 deprecated in the coming weeks. Teams with existing integrations should pin the version explicitly, re-run their own word-error-rate measurement against representative audio, and only then decide whether to move off 1.0 — rather than accepting the new default on the vendor's accuracy claim alone.

Sources: x.ai — Grok Voice Transcribe 2.0, MarkTechPost — coverage of the release.