CekuraBench
Book a demo

Google Chirp 2

Every modelGoogle

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 (%)

Google Chirp 2 is 7th of 19 on the Ocular Dataset and 16th of 16 on time to first text.

3.01% Overall WER (10th) · 2.59% Pipecat WER (11th) · 3.80% Ocular WER (7th) · 4.50s First text (16th) · 568ms Final text (16th)Likely range, 95% confidence: Pipecat WER 2.31 to 2.89%, Ocular WER 2.93 to 4.60%Published price $0.96 per hour of streamed audio. V2 Standard rate for the first 500,000 minutes a month; Chirp is billed as a Standard model. Source, checked Oct 8, 2026.

Last updated Methodology by Dileep Chagam

Google Chirp 2 against every other model

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

  1. 2.8%
    Transcribe Stream 1
  2. 2.9%
    Reson8
  3. 3.0%
    Nemotron 3.5 Transcribe
  4. 3.1%
    Ink 2
  5. 3.5%
    Pulse
  6. 3.5%
    Realtime Whisper
  7. 3.8%
    Chirp 2
  8. 3.9%
    STT-RT v5
  9. 3.9%
    Linden
  10. 4.2%
    Flux English
  11. 4.2%
    Chirp 3
  12. 4.2%
    Universal 3.5 Pro
  13. 4.4%
    Nova-3
  14. 4.5%
    Gradium
  15. 4.5%
    4o Mini Transcribe
  16. 4.5%
    Universal 3.6 Pro
  17. 4.5%
    Flux Multilingual
  18. 10.6%
    STT-1
  19. 12.5%
    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.
Ocular Dataset
8 recordings from 4 real phone conversations, licensed from Ocular.
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 Google Chirp 2 struggled

On real calls, dropped words are its biggest source of error. On clean clips, wrong words lead. For the wider field, see our pick of the best voice AI APIs for real-time audio.

45%

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

Median across all 19 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.

20ms100ms500ms2s
Typical, P50
568ms*
16th fastest of 18
Slow, P90
1.04s*
16th fastest of 18
Slowest, P95
1.22s*
15th fastest of 18

TTFT, first text

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

200ms1s5s20s
Typical, P50
4.50s
Slowest of 16
Slow, P90
6.40s
15th fastest of 16
Slowest, P95
8.15s
15th fastest of 16

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

* TTFS is measured after the stream closes, because this model has no controlled finish.

1,000 Pipecat clips · 8 Ocular recordings · 205 turns timed · 1 failed requests

Google Chirp 2 head to head

Every comparisonAll 19 modelsBenchmark source

Google Chirp 2 FAQ

  • Google Chirp 2 places highest on the Ocular Dataset, 7th of 19 at 3.80%. Its weakest placing is time to first text, 16th of 16 at 4.50s.

  • Google Chirp 2 transcribes the 1,000 public Pipecat clips at 2.59% word error rate and the eight licensed Ocular recordings at 3.80%. Word error rate counts incorrect, missing, and extra words against the reference transcript, so lower is better.

  • First text arrives after 4.50s at the median, and the transcript is final 568ms 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 eight licensed conversation recordings from Ocular for real speech. Latency comes from 206 streamed turns, using each provider's own supported finalization contract. Google Chirp 2 returned usable text for 1,000 of the public clips.

Building on Google? Test your agent the same way.

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

Book a demoStart free