ElevenLabs Scribe vs Deepgram Nova-3 vs Whisper
ElevenLabs’ Scribe v2 took the STT world by surprise — here is how it stacks against the incumbent Deepgram Nova-3 and open Whisper.
Verdict
Deepgram Nova-3 is the accuracy+price winner; Scribe v2 if you are in the ElevenLabs ecosystem.
Side-by-Side Comparison
| # | Model | WER (lower = better) | Speed (RTF) | First Partial | Cost / 1K Min | Mode |
|---|---|---|---|---|---|---|
| 2 | Scribe v2 ElevenLabs | 4.9% | 92x | 0.2s | $7.50 | Streaming & Batch |
| 1 | Nova-3 (Speech-to-Text) Deepgram | 4.8% | 110x | 0.18s | $4.30 | Streaming & Batch |
| 7 | Whisper large-v3 turbo OpenAI / Open Source | 5.4% | 85x | 0.35s | $6.00 | Batch Async |
Model-by-Model
Scribe v2 — ElevenLabs
Nova-3 still edges accuracy at 4.8% WER vs Scribe’s 4.9%, and it is cheaper: $4.30 vs $7.50 per 1K minutes.
Nova-3 (Speech-to-Text) — Deepgram
Scribe v2 (4.9% WER) brings ElevenLabs’ audio expertise to ASR with strong speaker diarization — worth it if you already use ElevenLabs TTS.
Whisper large-v3 turbo — OpenAI / Open Source
Whisper turbo remains the free-ish open-weights fallback at 5.4% WER.
FAQ
Is Scribe better than Whisper?
Yes — Scribe v2 scores 4.9% WER vs Whisper turbo’s 5.4%, and supports streaming. Whisper only wins on self-hosting and zero API cost.
Scribe vs Deepgram — which is better?
Nova-3 is marginally more accurate (4.8% vs 4.9% WER), faster to first partial (0.18s vs 0.20s) and 43% cheaper. Scribe counters with stronger diarization and the ElevenLabs stack.