xAI announced Grok Voice Transcribe 2.0, its newest speech-to-text model built on the audio foundation behind Grok Voice. Per the lab, internal evaluations show it is twice as accurate as Grok Voice Transcribe 1.0 while keeping the same price: $0.10 per hour of audio for batch and $0.20 for streaming.
Grok Voice already powers, xAI says, tens of thousands of customer-support calls a day, transcribes millions of hours of video narration, and runs voice agents in physical products, including the Grok assistant in Tesla vehicles. Training uses live, noisy, multilingual audio recorded across diverse environments.
Positioning emphasizes hard audio: flaky phone lines, competing voices, local accents and credentials read aloud. On the public Artificial Analysis leaderboard cited by xAI, Transcribe 2.0 ranked first for accuracy among 32 streaming models.
Internal evaluations cover 8 kHz telephony support audio, conversations with Grok, spoken credentials (numbers, emails, addresses) and short multilingual commands. On telephony, xAI claims to lead every model it tested; on short phrases across 19 languages, word error falls from 20.6% to 6.8% versus version 1.0.
Features include batch and streaming, word-level timestamps, speaker diarization at no extra cost, up to eight channels, biasing with up to 100 key terms, smart formatting of numbers and dates, filler removal and smart turn detection for voice agents. Existing Speech-to-Text API integrations get the accuracy lift with no code changes.
Atlassian Loom appears as a customer story: Sanchan Saxena, SVP of Teamwork Collection, highlighted accuracy for Loom video transcripts and closing the “context to code” loop by piping transcripts into developer tools. It is a concrete reason transcription matters for agents: audio becomes actionable text.
Matching 1.0 pricing is the commercial jab at rivals competing on milliseconds and cents per hour. xAI says 2.0 will soon be the API default and that 1.0 will be deprecated in coming weeks; teams that need the prior version can pin the model id.
In the autumn 2026 market, Microsoft is promoting MAI-Transcribe-2 with aggressive speed, accuracy and price claims, while OpenAI, Google and ElevenLabs hold established offers. Grok Voice Transcribe 2.0 bets on real telephony audio and the Grok/Tesla ecosystem as a data differentiator.
For agent teams, included diarization and turn detection reduce pieces to integrate. For compliance, formatting and filler controls let teams choose literalness versus readability by use case.
Neural Pulse sticks to what xAI’s announcement supports: accuracy gains at the same price, a cited leaderboard lead and production use cases. It does not treat those rankings as absolute truth or invent unpublished benchmarks.
The speech-to-text race has become strategic because voice agents need low latency and low error on proper names and codes. One fewer word-error point on credentials can be the difference between an automated flow and a human escalation.
With Transcribe 2.0, xAI strengthens its voice stack alongside the conversational Grok assistant. October 2026 makes clear that transcription is no longer a cheap afterthought: it is agent and productivity infrastructure.
