SpaceXAI Ships Grok Voice Transcribe 2.0 at the Same Price, Citing Half the Errors
On September 18, 2026, SpaceXAI released Grok Voice Transcribe 2.0, a speech-to-text model it says is twice as accurate as 1.0 at the same price. On its short-phrase set, word error rate falls from 20.6% to 6.8%. Batch stays $0.10 per hour and streaming $0.20. The API default slug is now grok-voice-transcribe-2.0.
TLDR
SpaceXAI, September 18, 2026: Grok Voice Transcribe 2.0, a speech-to-text model built on the audio foundation model behind Grok Voice. SpaceXAI says that across its real-world evaluations it is one of the most accurate transcription models available, and twice as accurate as Grok Voice Transcribe 1.0, at the same price. On the company’s short-phrase set, word error rate drops from 20.6% to 6.8%.
Price, unchanged: batch $0.10 per hour of audio, streaming $0.20 per hour, with diarization, timestamps, and key terms included. Existing Speech-to-Text API integrations are supposed to pick up the accuracy gain with no code changes. Docs list the default slug as grok-voice-transcribe-2.0 (1.0 still selectable), region us-east-1, batch and streaming, and keyterm prompting.
Independent ranking: SpaceXAI says that on the public Artificial Analysis leaderboard, Transcribe 2.0 ranks first for accuracy among 32 streaming models. That is the lab citing AA, not a number read off the live board in this pass. This is a transcription model, not a chat model, so text-arena Elo does not apply.
Product-line de-dupe: not Grok Voice Think Fast 2.0 (July 29 speech-to-speech). Not Grok 4.7 (September 21).
Why this story matters
Transcription price did not move. The claim that matters is the error cut on short phrases, and whether AA’s streaming board still shows that rank once other labs answer. Teams already on the 1.0 slug are the ones who feel it first, because the default changed underneath them.