Scribe v2
ElevenLabs' speech-to-text model — word timestamps, speaker labels, and audio-event tags for production transcription.
About
Scribe v2 is ElevenLabs' speech-to-text model, released November 2025 as the company's second-generation transcription engine. Beyond the transcript it returns word-level timestamps, speaker labels, and audio-event tags, with a usage object reporting audio seconds and dollar cost per request. A Scribe v2 Realtime variant (announced November 11, 2025) targets streaming use at roughly 150ms latency across 90+ languages for voice agents and live subtitling.
On the independent Hugging Face Open ASR Leaderboard it averages 5.83% word error rate in English — behind Cohere Transcribe (5.42%) and NVIDIA's Canary variant (5.63%), but ahead of OpenAI's Whisper Large v3 (7.44%). It is callable on OpenRouter as elevenlabs/scribe-v2 alongside the company's TTS models.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Independent: Hugging Face Open ASR Leaderboard — 5.83% avg WER (English), vs Cohere Transcribe 5.42% and Whisper Large v3 7.44%.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
API docs verified live (HTTP 200); independent leaderboard WER is published by Hugging Face, not the vendor. Docked for: (1) proprietary with no weights; (2) realtime-latency figures are provider-measured under optimized conditions.