Murf Finch-Global

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 (%)

Murf Finch-Global is 5th of 23 on final-text delay and 18th of 21 on time to first text.

3.33% Overall WER (16th) · 3.17% Pipecat WER (17th) · 3.62% Ocular WER (10th) · 1.83s First text (18th) · 51ms Final text (5th)Likely range, 95% confidence: Pipecat WER 2.84 to 3.51%, Ocular WER 2.73 to 4.49%Published price $0.12 per hour of streamed audio. Murf's published Speech to Text rate of $0.12 per hour of audio, billed per second. Source, checked Nov 13, 2026.

Last updated Methodology by Dileep Chagam

Murf Finch-Global against every other model

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

  1. 18ms
    Rumble-2
  2. 38ms
    STT-RT v5
  3. 41ms
    STT-1
  4. 47ms
    STT 3.7 Realtime Agent
  5. 51ms
    Finch-Global
  6. 57ms
    Scribe v3 Live
  7. 91ms
    Universal 3.6 Pro
  8. 103ms
    Nova-3
  9. 108ms
    Flux English
  10. 114ms
    Flux Multilingual
  11. 123ms
    Ink 2
  12. 123ms
    Transcribe Stream 1
  13. 147ms
    Linden
  14. 178ms
    Scribe Pro v2.1
  15. 180ms
    Universal 3.5 Pro
  16. 201ms
    Pulse
  17. 219ms
    Gradium
  18. 235ms
    Resonant-1
  19. 475ms
    Chirp 3
  20. 543ms
    Realtime Whisper
  21. 568ms
    Chirp 2
  22. 630ms
    4o Mini Transcribe
  23. 652ms
    4o Transcribe

NVIDIA Nemotron 3.5 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 Murf Finch-Global struggled

Wrong words are its biggest source of error on both real calls and clean clips.

42%

of its errors on the Ocular calls are wrong words.

Median across all 24 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
51ms
5th fastest of 23
Slow, P90
124ms
6th fastest of 23
Slowest, P95
155ms
9th fastest of 23

TTFT, first text

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

200ms1s5s20s
Typical, P50
1.83s
18th fastest of 21
Slow, P90
13.33s
Slowest of 21
Slowest, P95
16.14s
Slowest of 21

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

1,000 Pipecat turns · 300 Ocular turns · 205 turns timed · 0 failed requests

Murf Finch-Global FAQ

  • Murf Finch-Global places highest on final-text delay, 5th of 23 at 51ms. Its weakest placing is time to first text, 18th of 21 at 1.83s.

  • Murf Finch-Global transcribes the 1,000 public Pipecat clips at 3.17% word error rate and the 300 licensed Ocular turns at 3.62%. Word error rate counts incorrect, missing, and extra words against the reference transcript, so lower is better.

  • First text arrives after 1.83s at the median, and the transcript is final 51ms 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. Murf Finch-Global returned usable text for 1,000 of the public clips.

Building on Murf? Test your agent the same way.

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