Batch (pre-recorded)
We haven't written a description for this benchmark yet; see the card notes below for how each maker ran it.
- ModelScoreYour cost / mo
- 1Deepgram Nova-3 · Deepgram5.3%
“Best value”: no other model here reports a higher score at a lower cost for your workload. Scores are self-reported by each maker with different settings, so treat close calls as ties.
How each maker reported it
- Deepgram Nova-3 launch post
Deepgram's own evaluation: median word error rate (lower is better) on 2,703 files / 81.69 h across call, meeting, medical, drive-thru, ATC and other domains (the post says nine domains and lists ten). The competitor is not named in the text. Deepgram states 54.2% (streaming) and 47.4% (batch) lower WER than the next-best competitor. A separate preference study over 200+ samples says Nova-3 was preferred over OpenAI Whisper in all 7 languages tested, with no per-language figures.
Only each model's own maker's number is shown; scores a maker reports for rival models are left out.