- xAI describes v2.0 as twice as accurate as v1.0 on its internal sets
- Company claims v2.0 ranks first among 32 streaming models on a public leaderboard cited by the post
- Internal eval sets include telephony calls, Grok conversations, spoken credentials, and short multilingual commands
- Grok Voice Transcribe 2.0 will become the Speech-to-Text API default; v1.0 will be deprecated in coming weeks
- Customers can pin grok-voice-transcribe-1.0 if they want to remain on the earlier model
- Listed pricing: $0.10 per hour for batch, $0.20 per hour for streaming; diarization and timestamps included
- Supports automatic language detection, dozens of languages, speaker diarization, multichannel up to eight channels
- Includes word-level timestamps with confidence, key-term biasing, and smart turn detection
- Atlassian (Loom) is quoted as switching to v2.0 for video transcription in the announcement
Useful?