AssemblyAI STT 3.7 Realtime Agent

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 STT 3.7 Realtime Agent is 2nd of 24 on the Pipecat Dataset and 12th of 24 on the Ocular Dataset.

2.41% Overall WER (7th) · 1.68% Pipecat WER (2nd) · 3.81% Ocular WER (12th) · 493ms First text (4th) · 47ms Final text (4th)Likely range, 95% confidence: Pipecat WER 1.47 to 1.90%, Ocular WER 3.44 to 4.30%Published price $0.45 per hour of streamed audio. Pay as you go, billed while the session is open. Source, checked Nov 16, 2026.

Last updated Methodology by Dileep Chagam

AssemblyAI STT 3.7 Realtime Agent against every other model

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

  1. 1.5%
    Scribe Pro v2.1
  2. 1.7%
    STT 3.7 Realtime Agent
  3. 1.7%
    Rumble-2
  4. 1.8%
    Universal 3.6 Pro
  5. 1.8%
    Scribe v3 Live
  6. 1.9%
    Nemotron 3.5 Transcribe
  7. 1.9%
    Transcribe Stream 1
  8. 1.9%
    Universal 3.5 Pro
  9. 2.0%
    Chirp 3
  10. 2.1%
    Resonant-1
  11. 2.2%
    Realtime Whisper
  12. 2.4%
    Ink 2
  13. 2.5%
    Linden
  14. 2.5%
    STT-RT v5
  15. 2.6%
    Chirp 2
  16. 2.7%
    STT-1
  17. 3.2%
    Finch-Global
  18. 3.4%
    4o Mini Transcribe
  19. 3.5%
    Nova-3
  20. 3.7%
    Pulse
  21. 3.9%
    4o Transcribe
  22. 3.9%
    Flux English
  23. 4.0%
    Gradium
  24. 5.0%
    Flux Multilingual
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 STT 3.7 Realtime Agent struggled

On real calls, dropped words are its biggest source of error. On clean clips, wrong words lead.

67%

of its errors on the Ocular calls are words it left out.

Median across all 24 models: 51%

Mix of errors

Real calls (Ocular Dataset)

Clean clips (Pipecat Dataset)

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
47ms
4th fastest of 23
Slow, P90
102ms
5th fastest of 23
Slowest, P95
124ms
5th fastest of 23

TTFT, first text

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

200ms1s5s20s
Typical, P50
493ms
4th fastest of 21
Slow, P90
854ms
Fastest of 21
Slowest, P95
944ms
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 STT 3.7 Realtime Agent FAQ

  • AssemblyAI STT 3.7 Realtime Agent places highest on the Pipecat Dataset, 2nd of 24 at 1.68%. Its weakest placing is the Ocular Dataset, 12th of 24 at 3.81%.

  • AssemblyAI STT 3.7 Realtime Agent transcribes the 1,000 public Pipecat clips at 1.68% word error rate and the 300 licensed Ocular turns at 3.81%. Word error rate counts incorrect, missing, and extra words against the reference transcript, so lower is better.

  • First text arrives after 493ms at the median, and the transcript is final 47ms 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 STT 3.7 Realtime Agent 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.