OpenTranscription
OpenTranscription
RankerModelsPlayground
OpenTranscription
OpenTranscription

One API to every speech-to-text model worth using. Compare them on your audio, route to the best one, pay per second.

Platform status

Product

RankerModelsTranscriptionsPlaygroundBlog

Developers

DocumentationReliabilityAPI VersioningStatus

Legal

Privacy PolicyTerms of ServiceSupport
© 2026 OpenTranscription
Back to Models

Nova-2 vs GPT-4o Transcribe Diarize

2 models compared on the same benchmark — accuracy, latency, price, language coverage and capabilities. GPT-4o Transcribe Diarize leads on raw accuracy; GPT-4o Transcribe Diarize is fastest; Nova-2 covers the most languages; Nova-2 is cheapest.

Open both in playground

Verdict · who wins what

Most accurate

GPT-4o Transcribe Diarize

14.6% WER

OpenAI

Fastest

GPT-4o Transcribe Diarize

0.2× realtime

OpenAI

Most languages

Nova-2

33 languages

Deepgram

Cheapest

Nova-2

$0.0058/min

Deepgram

Comparing

2 models · same benchmark

1

Batch & Realtime

Nova-2

Deepgram

$0.0058/min
#18 · 76.2

Deepgram's Nova-2 speech recognition

Use This ModelDetails →

2

Batch

GPT-4o Transcribe Diarize

OpenAI

$0.0060/min
#12 · 50.7

GPT-4o transcription with built-in speaker diarization

Use This ModelDetails →

Overview

30-day benchmark average

Overall score

0–100

76.2

50.7

Accuracy rank

of 35

#18

#12

Word error rate

WER

16.5%

14.6%

Character error rate

CER

11.9%

10.0%

Match error rate

MER

15.4%

13.5%

Word info lost

WIL

21.0%

19.9%

Languages

33

12

Cost vs. accuracy

overall WER vs price

GPT-4o Transcribe Diarize

Nova-2

field · 33 models

lower-left is better

Accuracy by category

WER · lower is better

code_switching

63.7%
25.5%

conversational

10.2%
12.2%

finance

11.3%
6.3%

general

13.1%
14.5%

legal

3.8%
2.3%

medical

7.3%
1.9%

noisy

31.3%
32.5%

technical

4.9%
4.6%

Accuracy by accent

WER · English

african

26.2%
27.7%

australian

15.4%
29.2%

indian

25.4%
36.5%

uk

11.9%
13.4%

us

9.4%
18.9%

Speed & latency

p50 → p99

Processing speed

× realtime

0.1×

0.2×

Latency p50

1.6s

17.2s

Latency p90

3.7s

18.1s

Latency p99

3.7s

18.1s

Realtime / streaming

Streaming benchmark averages

Realtime rank

of 17

#4

—

Realtime score

0–100

78.0

—

Time to first word

986ms

—

Final drain

192ms

—

Real-time factor

1.02×

—

Flicker

10.5%

—

Cadence

0.8/s

—

Reliability

Rolling 30 days

Uptime

83.33%

100.00%

Error rate

16.67%

0.00%

Pricing

billed per second

Rate

per minute

$0.0058

$0.0060

Billing granularity

per 1s

per 1s

1 hour of audio

$0.35

$0.36

100 hours

$34.92

$36.00

1,000 hours

$349.20

$360.00

Capabilities

Feature support

Speaker diarization

Yes

Yes

Word-level timestamps

Yes

No

Code-switching

Yes

No

Auto-detect language

Yes

Yes

Live / streaming

Yes

No

Custom vocabulary

Yes

No

Coverage & deployment

Languages · formats · regions

Languages

33

12

Mode

Batch & Realtime

Batch

Region

United States

United States

Audio formats

mp3
wav
flac
m4a
ogg
webm
mp3
wav
flac
m4a
ogg
webm

Max file size

2 GB

25 MB

Max duration

No limit

No limit

Compliance

HIPAA
SOC2
GDPR
SOC2
GDPR

Frequently asked questions

Is Nova-2 or GPT-4o Transcribe Diarize more accurate?

GPT-4o Transcribe Diarize is more accurate, with a 14.6% word error rate versus 16.5% for Nova-2, on our standardized benchmark.

Which is cheaper, Nova-2 or GPT-4o Transcribe Diarize?

Nova-2 is cheaper at $0.0058/min versus $0.0060/min for GPT-4o Transcribe Diarize.

Should you choose Nova-2 or GPT-4o Transcribe Diarize?

Choose Nova-2 for lower cost and language coverage; choose GPT-4o Transcribe Diarize for accuracy and speed.