2 models compared on the same benchmark — accuracy, latency, price, language coverage and capabilities. Universal-3 Pro leads on raw accuracy; Universal-3 Pro is fastest; Nova-2 covers the most languages; Universal-3 Pro is cheapest.
Verdict · who wins what
Most accurate
Universal-3 Pro
17.3% WER
AssemblyAI
Fastest
Universal-3 Pro
0.2× realtime
AssemblyAI
Most languages
Nova-2
33 languages
Deepgram
Cheapest
Universal-3 Pro
$0.0035/min
AssemblyAI
Comparing
2 models · same benchmark
1
Universal-3 Pro
AssemblyAI
AssemblyAI's most powerful speech language model — up to 1000 keyterm phrases
2
Nova-2
Deepgram
Deepgram's Nova-2 speech recognition
30-day benchmark average
overall WER vs price
Nova-2
Universal-3 Pro
field · 29 models
lower-left is better
WER · lower is better
WER · English
p50 → p99
Streaming benchmark averages
Rolling 30 days
billed per second
Feature support
Languages · formats · regions
general
legal
medical
noisy
technical
uk
us
Latency p99
5.8s
1.6s
$20.88
$34.92
1,000 hours
$208.80
$349.20
Yes
Auto-detect language
Yes
Yes
Live / streaming
No
Yes
Custom vocabulary
Yes
Yes
Compliance
Is Universal-3 Pro or Nova-2 more accurate?
Universal-3 Pro is more accurate, with a 17.3% word error rate versus 31.4% for Nova-2, on our standardized benchmark.
Which is cheaper, Universal-3 Pro or Nova-2?
Universal-3 Pro is cheaper at $0.0035/min versus $0.0058/min for Nova-2.
Should you choose Universal-3 Pro or Nova-2?
Choose Universal-3 Pro for accuracy, speed, and lower cost; choose Nova-2 for language coverage.
