Sauti / deepgram
Deepgram Nova-3
Deepgram's entire pre-recorded lineup
WER
11.5%
95% CI 10.0%–13.1%
Rank
#2
of 12 measured
Latency
0.4s
mean, per clip
Cost
$0.46
per audio hour
Every other model on the board is a statistical tie with this one. Its confidence interval overlaps all 11 of them, so the rank above tells you where it landed, not that it is better or worse than anything else here.
What it is
Deepgram's general-purpose batch model, served over the /v1/listen HTTP endpoint. Their newer Flux generation is WebSocket-only conversational speech recognition on /v2/listen, so Nova-3 remains the batch product rather than a legacy one.
Pick it when
Latency is the standout, and it is not close: it was the fastest model on the board by a wide margin on identical audio, several times quicker than the next API model, while placing near the top on accuracy.
Think twice when
It is the most expensive model on this board pre-recorded, and multilingual costs more again. If cost matters more than milliseconds, models that tie it on accuracy are available for a fraction of the price.
Assessment written against the board of 2026-08-08. The figures above are read live from the current published snapshot (2026-08-08), so if those dates differ, trust the figures.
What the vendor publishes
deepgram publishes no stated word error rate figure for this model on any public benchmark. That absence is itself the finding. Source (read 2026-07-20).Deepgram states a RELATIVE reduction (~47% batch), not an absolute English WER. AssemblyAI measured it ~7.9% on their set.
How this was measured
250 clips (2.9 hours) of unscripted conversational English from People’s Speech (MLCommons, CC-BY), human-transcribed. Scored on a held-out split (N=77) never used to tune anything, through one fixed text normalizer, with bootstrap 95% confidence intervals. One corpus, one language, one modality: batch pre-recorded. It does not tell you how this model handles your accents, your domain vocabulary, or streaming.
Cost is deepgram’s own published pre-recorded pay-as-you-go list price, read from their pricing page on 2026-08-08. Streaming and committed-volume rates differ.Monolingual pre-recorded. Multilingual is $0.0092/min.