CekuraBench
Book a demo

Reson8 Rumble-2

Every model

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 (%)

Reson8 Rumble-2 is 1st of 21 on the Pipecat Dataset and 2nd of 18 on time to first text.

2.05% Overall WER (2nd) · 1.69% Pipecat WER (1st) · 2.75% Ocular WER (2nd) · 486ms First text (2nd) · 18ms Final text (1st)Likely range, 95% confidence: Pipecat WER 1.48 to 1.92%, Ocular WER 1.95 to 3.61%Published price $0.12 per hour of streamed audio. Reson8's real-time rate as of Oct 28, 2026. Source, checked Oct 8, 2026.

Last updated Methodology by Dileep Chagam

Reson8 Rumble-2 against every other model

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

  1. 1.7%
    Rumble-2
  2. 1.8%
    Universal 3.6 Pro
  3. 1.8%
    Scribe v3 Live
  4. 1.9%
    Nemotron 3.5 Transcribe
  5. 1.9%
    Transcribe Stream 1
  6. 1.9%
    Universal 3.5 Pro
  7. 2.0%
    Chirp 3
  8. 2.1%
    Resonant-1
  9. 2.2%
    Realtime Whisper
  10. 2.4%
    Ink 2
  11. 2.5%
    Linden
  12. 2.5%
    STT-RT v5
  13. 2.6%
    Chirp 2
  14. 2.7%
    STT-1
  15. 3.4%
    4o Mini Transcribe
  16. 3.5%
    Nova-3
  17. 3.7%
    Pulse
  18. 3.9%
    4o Transcribe
  19. 3.9%
    Flux English
  20. 5.0%
    Flux Multilingual
  21. 6.5%
    Gradium
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 Reson8 Rumble-2 struggled

On real calls, dropped words are its biggest source of error. On clean clips, wrong words lead. Our guide explains how word error rate is measured.

53%

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

Median across all 21 models: 49%

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
18ms
Fastest of 20
Slow, P90
35ms
Fastest of 20
Slowest, P95
40ms
Fastest of 20

TTFT, first text

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

200ms1s5s20s
Typical, P50
486ms
2nd fastest of 18
Slow, P90
983ms
3rd fastest of 18
Slowest, P95
987ms
2nd 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

Reson8 Rumble-2 head to head

Every comparisonAll 20 modelsBenchmark source

Reson8 Rumble-2 FAQ

  • Reson8 Rumble-2 places highest on the Pipecat Dataset, 1st of 21 at 1.69%. Its weakest placing is time to first text, 2nd of 18 at 486ms.

  • Reson8 Rumble-2 transcribes the 1,000 public Pipecat clips at 1.69% word error rate and the 300 licensed Ocular turns at 2.75%. Word error rate counts incorrect, missing, and extra words against the reference transcript, so lower is better.

  • First text arrives after 486ms at the median, and the transcript is final 18ms 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. Reson8 Rumble-2 returned usable text for 1,000 of the public clips.

Building on Reson8? Test your agent the same way.

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

Book a demoStart free