CekuraBench
All comparisons
GoogleOpenAI

Gemini 3.1 Flash Live Preview vs GPT Realtime 2.1 Mini

Gemini 3.1 Flash Live Preview is ahead on reliability, success rate, agent responses and data accuracy, and GPT Realtime 2.1 Mini on stalled calls, interruption, response time and cost.

Last updated Methodology by Dileep Chagam

Compare

Success rate by suite

Higher is better
Appointments59 scenariosMedicare23 scenarios
  • Gemini 3.1 Flash LiveGoogle
    93%
    32%
  • Realtime 2.1 MiniOpenAI
    84%
    4%
Each model's share of calls that passed every check, on each suite.
Verdict

Which to pick

Pick Gemini 3.1 Flash Live Preview for reliability across repeated calls (12.2 points higher), calls that pass every check (14.6 points higher), getting the agent's replies right (2.4 points higher) and saving the caller's data correctly (14.7 points higher). Pick GPT Realtime 2.1 Mini for fewer stalled calls (0.8 points lower), handling interruptions (0.01 higher), fast replies (0.80s sooner) and lower cost ($0.050 cheaper).

Every metric, side by side

Reliability

Scenarios that passed on all three runs.

Gemini 3.1 Flash Live
62.2%6th
Realtime 2.1 Mini
50.0%9th

Gemini 3.1 Flash Live, 12.2 points higher

Success rate

Calls that passed every check.

Gemini 3.1 Flash Live
76.0%5th
Realtime 2.1 Mini
61.4%10th

Gemini 3.1 Flash Live, 14.6 points higher

Agent responses

Calls where the agent's replies reached the expected outcome.

Gemini 3.1 Flash Live
95.9%6th
Realtime 2.1 Mini
93.5%10th

Gemini 3.1 Flash Live, 2.4 points higher

Data accuracy

Calls where everything the agent saved matched exactly.

Gemini 3.1 Flash Live
79.2%5th
Realtime 2.1 Mini
64.5%10th

Gemini 3.1 Flash Live, 14.7 points higher

Stalled calls

Calls where the agent went silent for 10 seconds or more.

Gemini 3.1 Flash Live
3.3%6th
Realtime 2.1 Mini
2.5%4th

Realtime 2.1 Mini, 0.8 points lower

Interruption

How well it handled being talked over, 0 to 5.

Gemini 3.1 Flash Live
4.985th
Realtime 2.1 Mini
4.992nd

Realtime 2.1 Mini, 0.01 higher

Response time

Median time to start replying after the caller finishes.

Gemini 3.1 Flash Live
2.88s11th
Realtime 2.1 Mini
2.08s8th

Realtime 2.1 Mini, 0.80s sooner

Cost / min

One minute of call at the vendor's published rates.

Gemini 3.1 Flash Live
$0.0703rd
Realtime 2.1 Mini
$0.0201st

Realtime 2.1 Mini, $0.050 cheaper

The small figure is the place among ranked models. The cascade is not ranked.

Realtime 2.1 Mini answers first

Realtime 2.1 Mini 2.08s, Gemini 3.1 Flash Live 2.88s at the median.

Response time, median to p90

Lower is betterMedianTo p90
Medianp90
  • Realtime 2.1 MiniOpenAI
    Realtime 2.1 Mini: median 2.08s, p90 2.75s
    2.08s2.75s
  • Gemini 3.1 Flash LiveGoogle
    Gemini 3.1 Flash Live: median 2.88s, p90 4.00s
    2.88s4.00s
2.0s3.0s4.0s
Seconds from the caller finishing to the agent starting to reply, per model.

What we saw on the calls

Gemini 3.1 Flash Live Preview

  • Reliable on bookings.
All 3 notes on Gemini 3.1 Flash Live
Model
models/gemini-3.1-flash-live-preview
Setting
Thinking level high
Voice
Charon
Turn taking
Shared local turn detection
Audio
16 kHz
Cost
Averaged over the calls with a usage record.

GPT Realtime 2.1 Mini

  • Gets most bookings right, at a fraction of the cost.
All 3 notes on Realtime 2.1 Mini
Model
gpt-realtime-2.1-mini
Setting
Reasoning high
Voice
marin
Turn taking
Service turn detection
Audio
24 kHz
Cost
Averaged over the calls with a usage record.
See the full leaderboardReference agent

Related comparisons

Both providers, every model: Google vs OpenAI

Frequently asked questions

Is Gemini 3.1 Flash Live Preview or GPT Realtime 2.1 Mini more reliable?

Gemini 3.1 Flash Live Preview is more reliable, by 12.2 points. Gemini 3.1 Flash Live Preview passed 51 of 82 scenarios on all three runs, and GPT Realtime 2.1 Mini passed 41. A scenario only counts when every run of it passed.

Which should I pick, Gemini 3.1 Flash Live Preview or GPT Realtime 2.1 Mini?

For a phone agent, Gemini 3.1 Flash Live Preview, since it gets more scenarios right on every run. Gemini 3.1 Flash Live Preview is ahead on reliability, success rate, agent responses and data accuracy, and GPT Realtime 2.1 Mini on stalled calls, interruption, response time and cost.

Which saves caller data more accurately, Gemini 3.1 Flash Live Preview or GPT Realtime 2.1 Mini?

Gemini 3.1 Flash Live Preview, by 14.7 points. Gemini 3.1 Flash Live Preview saved everything exactly right on 79.2% of the calls that save data, and GPT Realtime 2.1 Mini on 64.5%. On Appointments, 96.5% against 88.8%. On Medicare, 34.8% against 0.0%. A call counts only when nothing is missing, wrong, extra or saved twice.

Is Gemini 3.1 Flash Live Preview faster than GPT Realtime 2.1 Mini?

No. GPT Realtime 2.1 Mini starts replying 0.80s sooner at the median. Gemini 3.1 Flash Live Preview takes 2.88s at the median and 4.00s at p90; GPT Realtime 2.1 Mini takes 2.08s and 2.75s. Both are measured on the call audio, from the caller finishing to the agent starting to reply.

Which handles interruptions better, Gemini 3.1 Flash Live Preview or GPT Realtime 2.1 Mini?

GPT Realtime 2.1 Mini, by 0.01 on a 0 to 5 scale. Gemini 3.1 Flash Live Preview scores 4.98 and GPT Realtime 2.1 Mini 4.99 for how well the agent handled being talked over.

Which goes silent on callers less often, Gemini 3.1 Flash Live Preview or GPT Realtime 2.1 Mini?

GPT Realtime 2.1 Mini, by 0.8 points. The agent went silent for 10 seconds or more on 3.3% of Gemini 3.1 Flash Live Preview calls and 2.5% of GPT Realtime 2.1 Mini calls.

Which is cheaper, Gemini 3.1 Flash Live Preview or GPT Realtime 2.1 Mini?

GPT Realtime 2.1 Mini is $0.050 cheaper a minute. Gemini 3.1 Flash Live Preview costs $0.070 a minute of call and GPT Realtime 2.1 Mini $0.020, at each vendor's published rates.

Where does the Gemini 3.1 Flash Live Preview vs GPT Realtime 2.1 Mini data come from?

Cekura ran both as the whole agent in the same open-source Pipecat pipeline, with the same prompt, tools and connection. A simulated caller worked through all 82 scenarios on live calls, three times each, 2,952 calls across the benchmark. A call passes only when the agent said the right things and saved the right data.