CekuraBench
Book a demo

ElevenLabs Scribe v3 Live

Every modelElevenLabs

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 (%)

ElevenLabs Scribe v3 Live is 1st of 21 on both datasets together and 4th of 18 on time to first text.

1.89% Overall WER (1st) · 1.81% Pipecat WER (3rd) · 2.05% Ocular WER (1st) · 685ms First text (4th) · 57ms Final text (4th)Likely range, 95% confidence: Pipecat WER 1.59 to 2.04%, Ocular WER 1.57 to 2.52%Published price $0.39 per hour of streamed audio. Scribe v2 Realtime's published rate. ElevenLabs has not published a rate for Scribe v3 Live yet. Source, checked Nov 10, 2026.

Last updated Methodology by Dileep Chagam

ElevenLabs Scribe v3 Live against every other model

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

  1. 1.9%
    Scribe v3 Live
  2. 2.1%
    Rumble-2
  3. 2.2%
    Transcribe Stream 1
  4. 2.3%
    Nemotron 3.5 Transcribe
  5. 2.4%
    Resonant-1
  6. 2.6%
    Ink 2
  7. 2.6%
    Realtime Whisper
  8. 2.7%
    Universal 3.6 Pro
  9. 2.7%
    Universal 3.5 Pro
  10. 2.8%
    Chirp 3
  11. 3.0%
    Linden
  12. 3.0%
    Chirp 2
  13. 3.0%
    STT-RT v5
  14. 3.6%
    Pulse
  15. 3.7%
    4o Mini Transcribe
  16. 3.8%
    Nova-3
  17. 4.0%
    Flux English
  18. 4.8%
    Flux Multilingual
  19. 5.4%
    STT-1
  20. 5.8%
    Gradium
  21. 6.9%
    4o Transcribe
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 ElevenLabs Scribe v3 Live struggled

Wrong words are its biggest source of error on both real calls and clean clips. For more on the platform, read our ElevenLabs review.

44%

of its errors on the Ocular calls are wrong words.

Median across all 21 models: 34%

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
57ms
4th fastest of 20
Slow, P90
83ms
4th fastest of 20
Slowest, P95
89ms
4th fastest of 20

TTFT, first text

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

200ms1s5s20s
Typical, P50
685ms
4th fastest of 18
Slow, P90
1.02s
4th fastest of 18
Slowest, P95
1.32s
6th fastest of 18

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

1,000 Pipecat turns · 300 Ocular turns · 205 turns timed · 1 failed requests

ElevenLabs Scribe v3 Live head to head

Every comparisonAll 20 modelsBenchmark source

ElevenLabs Scribe v3 Live FAQ

  • ElevenLabs Scribe v3 Live places highest on both datasets together, 1st of 21 at 1.89%. Its weakest placing is time to first text, 4th of 18 at 685ms.

  • ElevenLabs Scribe v3 Live transcribes the 1,000 public Pipecat clips at 1.81% word error rate and the 300 licensed Ocular turns at 2.05%. Word error rate counts incorrect, missing, and extra words against the reference transcript, so lower is better.

  • First text arrives after 685ms at the median, and the transcript is final 57ms 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. ElevenLabs Scribe v3 Live returned usable text for 1,000 of the public clips.

Building on ElevenLabs? Test your agent the same way.

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

Book a demoStart free