xAI releases Grok Voice Transcribe 2.0 speech-to-text model

xAI released Grok Voice Transcribe 2.0, its latest speech-to-text model, on September 18, 2026, holding batch pricing at $0.10 per hour of audio while describing the model as twice as accurate as Grok Voice Transcribe 1.0.

Grok Voice Transcribe 2.0 is built on the audio foundation model behind Grok Voice. According to xAI, Grok Voice already powers 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. The company said the new model was trained on live, noisy, multilingual audio recorded across a diverse set of environments and refined with post-training, and it described the result as one of the most accurate transcription models available for speech in real-world settings.

Accuracy Evaluations

xAI said Grok Voice Transcribe 2.0 ranks first for accuracy among 32 streaming models on the public Artificial Analysis leaderboard. Beyond public benchmarks, the company measures word error rate on four internal evaluation sets drawn from production traffic: telephony audio from customer-support calls, conversations with Grok, spoken credentials such as account codes and email addresses, and short multilingual voice commands. The company reported that the new model improves on Grok Voice Transcribe 1.0 across all four sets and leads every model it tested on the telephony set, which consists of 8 kHz English customer-support calls. xAI’s published charts compare the model against Gemini 3.5 Transcribe, MAI-Transcribe-2, ElevenLabs Scribe v2, Deepgram Nova-3, and Whisper Large v3 on those internal sets.

Multilingual transcription is the largest accuracy gain over version 1.0, according to xAI. The company said the model transcribes dozens of languages, detects the language automatically, and follows mid-recording language switches in a single pass. Short phrases, such as in-car commands, leave the model little context from which to identify the language; on xAI’s short-phrase set of voice-assistant utterances spanning 19 languages, word error rate drops from 20.6 percent to 6.8 percent, the company reported.

Features and API Access

Through the Speech-to-Text API, Grok Voice Transcribe 2.0 handles batch transcription of recorded files and URLs as well as real-time streaming. The documented feature set includes word-level timestamps with confidence scores, speaker diarization at no additional cost, multichannel transcription of up to eight channels, key term biasing of up to 100 domain terms per request, text formatting that returns numbers, dates, currencies, phone numbers, and email addresses in written form, filler word removal, and smart turn detection that identifies the end of a speaker’s turn for voice agents. xAI said existing Speech-to-Text API integrations receive the accuracy improvement with no code changes.

The official speech-to-text documentation lists 12 supported audio formats, a maximum file size of 500 MB, and sample rates of 8000, 16000, 22050, 24000, 44100, and 48000 Hz. A language parameter enables written-form formatting across 25 languages, among them English, Spanish, French, German, Hindi, Japanese, and Korean.

Batch requests use multipart form data and must supply either an uploaded file or a URL for the server to download and transcribe; the response returns the full transcript, the detected language as a BCP-47 code, audio duration in seconds, and word-level segments with start and end times. For streaming, clients send raw audio as binary frames to a WebSocket endpoint at wss://api.x.ai/v1/stt and receive JSON transcript events as the audio is processed, with optional interim results emitted roughly every 500 milliseconds.

Loom Deployment, Pricing, and Deprecation

xAI said Atlassian evaluated Grok Voice Transcribe 2.0 against its existing transcription solution, found it more accurate, and now uses the model to transcribe every video on Loom, its screen-recording product. xAI’s announcement quoted Sanchan Saxena, senior vice president of Teamwork Collection at Atlassian, on workflows that pipe Loom transcripts into the Cursor coding tool: “With Grok powering Loom’s speech-to-text and Cursor turning that into code, we’re closing the loop from context to code: record what you mean, and the work gets done.”

Pricing is identical to Grok Voice Transcribe 1.0: $0.10 per hour of audio for batch transcription and $0.20 per hour for streaming, with diarization, timestamps, and key terms included. xAI said Grok Voice Transcribe 2.0 will soon become the default model in the Speech-to-Text API and that version 1.0 will be deprecated in the coming weeks; customers that want to remain on the earlier model during the transition can pin grok-voice-transcribe-1.0 in their requests. The documentation currently lists grok-voice-transcribe-1.0 as the default when the model parameter is omitted, with grok-voice-transcribe-2.0 selectable on both the REST and WebSocket endpoints.