xAI ships Grok Voice Transcribe 2.0
xAI has released Grok Voice Transcribe 2.0, a speech-to-text model it says is twice as accurate as its predecessor at unchanged pricing and with no code changes required for existing API integrations.
xAI released Grok Voice Transcribe 2.0, a speech-to-text model it says is twice as accurate as its predecessor at the same price.
The model is built on the audio foundation model behind Grok Voice, which xAI says already handles tens of thousands of customer-support calls a day and powers the Grok assistant in Tesla vehicles. xAI says training used live, noisy, multilingual audio recorded across varied environments, then post-training refinement.
On the public Artificial Analysis leaderboard, xAI says the model ranks first for accuracy among 32 streaming models. Beyond that, the company reports word error rate on four internal sets drawn from production traffic — telephony support calls, conversations with Grok, spoken credentials such as account codes and email addresses, and short voice-assistant utterances across 19 languages. It claims gains on all four over version 1.0 and the lowest error rate of any model it tested on telephony. The single figure disclosed: on the short-phrase set, WER falls from 20.6% to 6.8%.
Pricing is unchanged — $0.10 per hour of audio for batch, $0.20 for streaming, with diarization, timestamps, and key terms included. Existing Speech-to-Text API integrations get the improvement with no code changes. Atlassian found the model more accurate than its prior solution and now uses it to transcribe every Loom video.
Grok Voice Transcribe 2.0 will soon become the API default; 1.0 is deprecated in the coming weeks and can be pinned as grok-voice-transcribe-1.0 during the transition. xAI did not publish the underlying WER tables or leaderboard scores in the announcement.