xAI releases Grok Voice Transcribe 2.0, twice as accurate at the same price

xAI released Grok Voice Transcribe 2.0, built on the audio foundation model behind Grok Voice, which already powers tens of thousands of customer-support calls daily, transcribes millions of hours of video narration, and runs voice agents in Tesla vehicles. The company says it is twice as accurate as Grok Voice Transcribe 1.0 at identical pricing — $0.10/hour batch, $0.20/hour streaming — and ranks first for accuracy among 32 streaming models on the public Artificial Analysis leaderboard.

The model was tuned for hard, real-world audio rather than clean single-speaker clips: flaky phone lines, competing voices, local accents, and spoken credentials like account codes and email addresses. On xAI's internal telephony set drawn from production customer-support traffic, it leads every model tested. Multilingual accuracy is the largest improvement — on short in-car command phrases, where the model has little context to identify the language, word error rate drops from 20.6% to 6.8%. It detects language automatically and follows mid-recording language switches in a single pass.

The release ties into a broader xAI voice push: the company also rolled out voice mode across Grok Bot on desktop and mobile, powered by Grok Voice Think Fast 2.0. Elon Musk, meanwhile, mapped a roadmap to AGI — announcing a 2.5-trillion-parameter Grok 4.8 in training and framing Grok 5 as the AGI milestone — days after appearing to endorse an AI slowdown. Competitively, undercutting incumbents like OpenAI's Whisper-class and Deepgram on accuracy-per-dollar for noisy production audio is a sharp enterprise wedge; the Tesla in-car deployment gives xAI a real-world proving ground rivals lack.