xAI Launches Grok Voice Transcribe 2.0, Tops Streaming ASR Leaderboard
On Sept 18-19, xAI released Grok Voice Transcribe 2.0, billing it as "the world's most accurate speech transcription model," claiming roughly 2x accuracy over v1 across customer-service calls, dictated credentials, and short voice commands at unchanged pricing. Artificial Analysis subsequently confirmed it tops the streaming transcription leaderboard, and Atlassian has integrated it into Loom. The combination of top streaming accuracy and low pricing pressures existing players.
Confirmed
- Built on the Grok Voice audio foundation model; official claim of 2x accuracy at unchanged prices
- Pricing: $0.10/hour batch, $0.20/hour streaming ($3.33 per 1,000 minutes per Artificial Analysis, slightly above Muse Voice Transcribe's $3, below Cartesia, roughly half of some competitors)
- Artificial Analysis (09-19): streaming final transcript AA-WER of 2.7%, first place among 32 models, final transcript delivered 0.49s after speech ends; first partial transcript 3.4% WER @0.49s — per @XFreeze, first in partial-transcript accuracy but slower than Muse Voice
- Non-streaming AA-WER of 2.3% (vs 4.0% for v1), 5th among 59 models, behind Alibaba's Fun-Realtime and others
- Atlassian has integrated the model in Loom
Unconfirmed
- @XFreeze's figures of 92.9% (phone audio) and 97.4% (overall) accuracy were labeled as claimed, with methodology unclear and differing from AA's WER metric
- @haider1 noted limited disclosure of benchmarks, pricing, and availability at posting time (later partially filled by AA's evaluation)
Why it matters
- Topping the streaming leaderboard at $0.10/hour batch pricing puts dual price-and-performance pressure on Deepgram, ElevenLabs, Gemini, Cartesia and others
- Not winning non-streaming (5th) and slower first partial transcripts show its edge is concentrated in streaming final transcription, not a clean sweep
- The Atlassian/Loom integration signals real production deployment; watch for further third-party validation and competitor price responses
2026-09-18 ~ 2026-09-19 · 15 related posts
Primary sources
- xAI launches Grok Voice Transcribe 2.0, tops 32 streaming models at $0.10/hr — SpaceXAI ·
- Grok Voice Transcribe 2.0 Tops AA-WER Streaming With 2.7% WER, Edging Out ElevenLabs and Cartesia — ArtificialAnlys ·
- xAI launches Grok Voice Transcribe 2.0, claiming 2x accuracy in calls and voice commands — SpaceXAI ·
- Grok Voice Transcribe 2.0 debuts at 97.4% accuracy, topping ElevenLabs and Gemini at $0.10/hour — XFreeze · 2026-09-18
- [source] xAI launches Grok Voice Transcribe 2.0, claiming 2x accuracy in calls and voice commands — SpaceXAI · 2026-09-19
- Grok launches Voice Transcribe 2.0, claims world's most accurate speech transcription — haider1 · 2026-09-19
- Grok Voice Transcribe 2.0 tops streaming ASR accuracy with 2.7% WER at same price — ArtificialAnlys · 2026-09-19
- [source] Grok Voice Transcribe 2.0 Tops AA-WER Streaming With 2.7% WER, Edging Out ElevenLabs and Cartesia — ArtificialAnlys · 2026-09-19
- Grok Voice Transcribe 2.0 hits 3.4% WER on first partial transcript, trading speed for accuracy — ArtificialAnlys · 2026-09-19
- Grok Voice Transcribe 2.0 scores 2.3% WER non-streaming, #5 of 59 models — ArtificialAnlys · 2026-09-19
- Grok Voice Transcribe 2.0 pricing: $3.33 per 1,000 streaming minutes, about half rivals — ArtificialAnlys · 2026-09-19
- Grok Voice Transcribe 2.0 Reported: 2x Accuracy, #1 Among 32 Streaming STT Models (Unverified) — NicoVerderosa · 2026-09-19
6 near-duplicate retellings: XFreeze · SpaceXAI · XFreeze · minchoi · XFreeze · emmanuelvivier