AssemblyAI Universal 3.6 Pro

Accuracy vs. speed

Lower and further left is better · Nemotron 3.5 Transcribe doesn't support streaming, so it is not shown

Best quadrantBest picks
Final-text delay P50 (ms)
Overall WER (%)

AssemblyAI Universal 3.6 Pro is 1st of 21 on time to first text and 21st of 24 on the Ocular Dataset.

2.71% Overall WER (10th) · 1.77% Pipecat WER (4th) · 4.50% Ocular WER (21st) · 485ms First text (1st) · 91ms Final text (7th)Likely range, 95% confidence: Pipecat WER 1.55 to 2.00%, Ocular WER 3.66 to 5.34%Published price $0.45 per hour of streamed audio. Pay as you go, billed while the session is open. Source, checked Oct 8, 2026.

Last updated Methodology by Dileep Chagam

AssemblyAI Universal 3.6 Pro against every other model

Lower is better on every measure. Hover a column for that model's full numbers.

  1. 485ms
    Universal 3.6 Pro
  2. 486ms
    Rumble-2
  3. 489ms
    Universal 3.5 Pro
  4. 493ms
    STT 3.7 Realtime Agent
  5. 685ms
    Scribe v3 Live
  6. 730ms
    Transcribe Stream 1
  7. 791ms
    Flux English
  8. 793ms
    Flux Multilingual
  9. 793ms
    Scribe Pro v2.1
  10. 1.06s
    Pulse
  11. 1.06s
    STT-1
  12. 1.06s
    Nova-3
  13. 1.19s
    Linden
  14. 1.20s
    STT-RT v5
  15. 1.38s
    Realtime Whisper
  16. 1.51s
    Ink 2
  17. 1.78s
    Gradium
  18. 1.83s
    Finch-Global
  19. 2.40s
    Resonant-1
  20. 4.48s
    Chirp 3
  21. 4.50s
    Chirp 2

NVIDIA Nemotron 3.5 Transcribe and GPT-4o Mini Transcribe and GPT-4o Transcribe send no text until the speaker stops, so TTFT is not comparable.

Glossary: WER, TTFT, TTFS, P50
WER · Word error rate
Share of words the model got wrong: replaced, missing or added. Lower is better.
Pipecat Dataset
1,000 short public clips from Pipecat's STT benchmark dataset. Each clip is one turn.
Ocular Dataset
300 turns from 4 real phone conversations, licensed from Ocular. 206 of them are timed.
Likely range · 95% range
Where the model's true score probably falls, given how much audio was tested. Overlapping ranges are too close to call.
TTFT · Time to first text
How long after the audio starts streaming until the first words appear.
TTFS · Final-text delay
How long after the caller stops until the full transcript arrives. A voice agent waits for this before it replies.
P50 / P90 / P95 · Typical, slow and slowest turns
Half of turns are faster than P50. Only 1 in 10 is slower than P90, and 1 in 20 slower than P95.

Where AssemblyAI Universal 3.6 Pro struggled

On real calls, dropped words are its biggest source of error. On clean clips, wrong words lead. For costs, see our breakdown of AssemblyAI pricing.

Wrong words

What the speaker said, with AssemblyAI Universal 3.6 Pro's mistakes marked: I just received a suspicious text message that claims my bank account has been locked due to some suspicious activity and it includes a strange link but I'm hesitant written as I mess it up

Everything AssemblyAI Universal 3.6 Pro wrote

I just received a suspicious text message that claims my bank account has been locked due to some suspicious activity. And it includes a strange link. But I mess it up.

Pipecat Dataset clip pipecat-ee49bba6-ad9f-6328-8d52-87700558bc9f

Wrong words

What the speaker said, with AssemblyAI Universal 3.6 Pro's mistakes marked: Please add organic kale written as Gala, apples, 2% whole milk, and a loaf of sourdough written as sort of bread to my main shopping list

Everything AssemblyAI Universal 3.6 Pro wrote

Please add organic Gala apples, 2% whole milk, and a loaf of sort of bread to my main shopping list.

Pipecat Dataset clip pipecat-e531272b-cdf6-5862-9400-1f439f03af13

Marked: words AssemblyAI Universal 3.6 Pro left out or got wrong. Audio from the Pipecat Dataset, under CC BY 4.0, re-encoded as MP3.

TTFT and TTFS at P50, P90 and P95

Timed on the same 206 streamed Ocular turns as every other model. P90 and P95 show what a caller meets on a slow turn.

TTFS, final text

How long after the caller stops until the full transcript arrives. A voice agent waits for this before it replies.

10ms50ms200ms1s
Typical, P50
91ms
7th fastest of 23
Slow, P90
210ms
13th fastest of 23
Slowest, P95
232ms
12th fastest of 23

TTFT, first text

How long after the audio starts streaming until the first words appear.

200ms1s5s20s
Typical, P50
485ms
Fastest of 21
Slow, P90
966ms
2nd fastest of 21
Slowest, P95
967ms
2nd fastest of 21

Every other model's typical turn, on the same axis.

1,000 Pipecat turns · 300 Ocular turns · 206 turns timed · 0 failed requests

AssemblyAI Universal 3.6 Pro FAQ

  • AssemblyAI Universal 3.6 Pro places highest on time to first text, 1st of 21 at 485ms. Its weakest placing is the Ocular Dataset, 21st of 24 at 4.50%.

  • AssemblyAI Universal 3.6 Pro transcribes the 1,000 public Pipecat clips at 1.77% word error rate and the 300 licensed Ocular turns at 4.50%. Word error rate counts incorrect, missing, and extra words against the reference transcript, so lower is better.

  • First text arrives after 485ms at the median, and the transcript is final 91ms after the speaker stops. Both are measured on the same 206 streamed turns as every other model, and the page shows P90 and P95 as well, which is what a caller meets on a bad turn.

  • Every model runs the same two datasets under the same conditions: 1,000 public clips from Pipecat's STT benchmark dataset for accuracy at scale, and 300 turns of licensed Ocular conversations for real speech. Latency comes from 206 streamed turns, using each provider's own supported finalization contract. AssemblyAI Universal 3.6 Pro returned usable text for 1,000 of the public clips.

Building on AssemblyAI? Test your agent the same way.

Cekura runs these simulated calls on your agent and scores every one.