Scribe v2
ElevenLabs' speech-to-text model: 99 languages, realtime streaming under 150 ms, and the cheapest published realtime rate in its class.
About
Scribe v2 is ElevenLabs' entry into speech recognition, and it arrived from the opposite direction of most ASR vendors: the company built its name on voice synthesis, then bolted recognition on. It covers 99 languages with automatic language detection, character- and word-level timestamps, speaker diarization, audio event tagging for non-speech sounds, and entity detection across dozens of categories for PII redaction. Keyterm prompting lets teams feed it up to 100 domain terms so product names and technical jargon land correctly.
Scribe v2 Realtime is the streaming variant, served over WebSocket with a vendor-claimed first-partial latency near 150 milliseconds across 90+ languages, plus voice activity detection and negative-latency word prediction. Independent Artificial Analysis testing in July 2026 put Scribe v2 at 2.2% AA-WER for batch and the realtime variant at 3.6% for streaming, leading both boards. Batch runs $0.22 per hour and realtime $0.39 per hour at flat per-plan rates, the cheapest published realtime rate in its class, though the architecture behind the model is undisclosed.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- Artificial Analysis AA-WER: 2.2% non-streaming (batch), 3.6% streaming (realtime) - leading both July 2026 boards (independent)
- Lowest WER recorded on FLEURS and Common Voice (provider-claimed, no published numbers)
- ~150 ms first-partial latency on realtime (provider-claimed)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 9/10
Scored 9/10 (verified tier, metrics-v1). Evidence: $0.22/hr batch and $0.39/hr realtime pricing confirmed on ElevenLabs' own pricing page (fetched Aug 2026) and in two independent API comparisons; 2.2% AA-WER batch and 3.6% streaming figures from Artificial Analysis' July 2026 boards; 99 languages and sub-150 ms latency from the Scribe v2 Realtime launch materials. Strengths: independent benchmark leadership on both batch and streaming, cheapest published realtime rate, strong enterprise posture (SOC 2, HIPAA, zero-retention mode). Docked for: (1) weights and architecture are closed, so accuracy claims rest on trust rather than reproduction; (2) the vendor's FLEURS/Common Voice 'lowest WER' claim has no published numbers. License: proprietary (API-only).
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 9.0/10
- Maintenance · recency, releases, issue load
- 9.0/10
- Openness · license + docs
- 2.0/10
- Manual review · sanity read + red-flag check
- 9.0/10
- · $0.22/hr batch, $0.39/hr realtime from elevenlabs.io/pricing/api (Aug 2026); 99 languages, sub-150ms realtime latency
- · AA-WER 2.2% batch / 3.6% streaming (Artificial Analysis, Jul 2026, independent); vendor FLEURS/Common Voice claim has no published numbers
- · Free plan: 10,000 monthly credits (STT at 330 credits/min)