Deepgram vs AssemblyAI: Speech-to-Textlatency & accuracy
Measured head-to-head, not marketing: the verdict is split — Deepgram is faster (98 ms median TTFS vs 363 ms), while AssemblyAI is more accurate (2.5% WER vs 4.5%). Independent data, refreshed daily, on identical inputs. Voice quality and price are not measured.
Deepgram or AssemblyAI — which is better?
Each provider's best model on each measured axis, last 7d. Lower is better on both:
| Axis | Deepgram (best model) | AssemblyAI (best model) | Measured winner |
|---|---|---|---|
| Speed — median TTFS | nova-3 · 98 ms | universal-3-pro · 363 ms | Deepgram |
| Accuracy — avg WER | nova-3 · 4.5% | universal-3.5-pro · 2.5% | AssemblyAI |
Not measured here: price — treat those as vendor claims until measured. Full leaderboard context: the independent speech-to-text benchmark →
Every Deepgram and AssemblyAI model, measured
All benchmarked models from both providers on the same pinned dataset, ranked fastest-first (median TTFS), last 7d:
| Model | Provider | TTFS median | TTFS p95 | WER | Samples |
|---|---|---|---|---|---|
| nova-3 | Deepgram | 98 ms | 154 ms | 4.5% | 3,327 |
| nova-2 | Deepgram | 98 ms | 175 ms | 5.2% | 3,330 |
| universal-3-pro | AssemblyAI | 363 ms | 883 ms | 3.1% | 650 |
| universal-3.5-pro | AssemblyAI | 419 ms | 663 ms | 2.5% | 3,360 |
Independent data, by Coval
Numbers are from the voice benchmark by Coval, a voice-AI evaluation platform that does not sell STT models — it measures every provider as a neutral third party on a pinned dataset under production-realistic conditions. Openbenchmarks mirrors the results daily with attribution; methodology and runner are open-source and reproducible.
Deepgram vs AssemblyAI — common questions
Is Deepgram faster than AssemblyAI for speech-to-text?
On the current 7d window, Deepgram's fastest model (nova-3) has a median TTFS of 98 ms, vs 363 ms for AssemblyAI's fastest (universal-3-pro) — so Deepgram is faster on measured median latency. Distributions matter too: p95 is 154 ms for Deepgram vs 883 ms for AssemblyAI.
Which is more accurate, Deepgram or AssemblyAI?
By Word Error Rate: Deepgram's best model (nova-3) averages 4.5%, vs 2.5% for AssemblyAI's best (universal-3.5-pro) — AssemblyAI leads on measured accuracy. Lower is better; WER is measured on identical inputs under production-realistic conditions.
Which should I pick — Deepgram or AssemblyAI?
The measured verdict is split: Deepgram is faster (median TTFS), AssemblyAI is more accurate (WER). Pick by your constraint — real-time voice agents care about latency; fidelity-critical workloads care about WER. Note price is not measured here.
Where does this data come from?
The independent voice benchmark by Coval, a voice-AI evaluation platform that does not sell STT models. Every model runs the same pinned dataset under production-realistic conditions, re-run roughly every 30 minutes; Openbenchmarks mirrors the results daily with attribution. Methodology and runner are open-source (Apache-2.0).