CekuraBench

Gemini 3.8 Live

Every modelCompare

41.5%

Reliability

12th of 12

1.91s

Median reply

5th of 12

Reliability vs. response time

Higher and further left is better

Best quadrant
Reliability, pass³ (%)
Median response time (s)

For the wider field, see our pick of the best voice AI APIs for real-time audio.

Gemini 3.8 Live against every other model

Scenarios that passed on all 3 runs, pass³. Higher is better.

#ModelReliability
  1. 01Grok Voice 3.081.7%
  2. 02GPT Realtime 2.179.3%
  3. 03GPT-Live 1 + GPT-6 Sol75.6%
  4. 04Azure Realtime70.7%
  5. 05Phonic v168.3%
  6. 06Grok Voice Think Fast 2.067.1%
  7. 07Gemini 3.1 Flash Live Preview62.2%
  8. 08Deepslate Opal57.3%
  9. 09Gemini 3.8 Live Extended Thinking52.4%
  10. 10Amazon Nova 2 Sonic50.0%
  11. 11GPT Realtime 2.1 Mini50.0%
  12. 12Gemini 3.8 Live41.5%
  • refCascade82.9%

Where Gemini 3.8 Live struggled

Gemini 3.8 Live falls furthest behind on caller corrects or changes mind: 38% passed, against a median of 81%.

Below the medianPassedMedian
  • Caller corrects or changes mind
    38%median 81%
  • Accent & slow speech
    50%median 83%
  • Background noise
    62%median 81%
  • Several requests in one call
    86%median 95%
  • Silence, cough & packet loss
    79%median 88%
  • Plain booking & lookup
    87%median 94%
  • Interruptions & barge-in
    87%median 90%

Response time at P50, P90 and P95

Gemini 3.8 Live starts replying in 1.91s on a typical turn and 2.43s on the slowest turn in twenty, 5th fastest of 12 at the median.

ModelP50P90P95
  • Cascade
    2.17s2.51s2.62s
Seconds from the caller finishing to the agent replying. The bar is the typical turn, P50.

Gemini 3.8 Live head to head

Every comparisonAll 12 models

How much audio Gemini 3.8 Live was tested on

246 live calls: 82 scenarios, each run three times.

Live calls
246
Scenarios
82
Booking calls
177
Medicare calls
69
Replies per call
6.0

82 scenarios, the same for every model. See every test case

Gemini 3.8 Live FAQ

  • Gemini 3.8 Live passed 34 of 82 scenarios on all three runs, 41.5%, 12th of 12 realtime models. Across all 246 calls, 69.1% passed every check.

  • Gemini 3.8 Live starts replying 1.91s after the caller finishes at the median, 2.23s at P90 and 2.43s at P95. That is 5th fastest of 12 at the median, measured on the call audio.

  • Gemini 3.8 Live falls furthest behind the field on caller corrects or changes mind: 38% of calls passed, against a median of 81%.

  • Gemini 3.8 Live saved everything exactly right on 70.5% of the calls that save data, 10th of 12. On Booking calls, 78.4%. On Medicare calls, 50.0%.

  • There is no verified per-minute rate for Gemini 3.8 Live, so it is left out of the cost ranking.

  • Cekura ran Gemini 3.8 Live as the whole agent in the same open-source Pipecat pipeline as every other model, with the same prompt, tools and connection. A simulated caller worked through 82 scenarios on live calls, three times each: 246 calls for Gemini 3.8 Live. A call passes only when the agent said the right things and saved the right data.