Benchmark Almanac

Gemini 3.1 Flash Live

Gemini 3.1 Flash Live ranks first on Speko Speech-to-Speech Benchmark (10-turn call).

17
Benchmarks with a score
1
Ranked first, out of five or more models
4
In the top three
Google DeepMind Mar 2026

Gemini 3.1 Flash Live

Closed weights Mid-size

No Epoch Capabilities Index score yet
  • Audio MC 36.1% #10 /21
  • τ-Voice 43.9% #5 /8
  • Realtime-Bench 13.3% #8 /11

Where has it been number one?

Gemini 3.1 Flash Live is number one on 1 benchmark today. It also led 4 more before newer models passed it.

How does it do against its rivals?

Gemini 3.1 Flash Live beats Grok Voice 2.0 on 3 of the 8 benchmarks they share. GPT-Realtime-2.1 does best against it, ahead on 6 of 9.

  1. Grok Voice 2.0 3 ahead, 5 behind Compare
  2. GPT-Realtime-2.1 2 ahead, 6 behind, 1 tied Compare

Sources: Official leaderboards, Epoch AI (CC BY 4.0) and labs’ model cards

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Agents and tools

BenchmarkScoreRankSource
Speko Speech-to-Speech Benchmark (10-turn call) 1108% #1 of 6 Official leaderboard
Realtime-Bench 13.3% #8 of 11 Official leaderboard
Speko Speech-to-Speech Benchmark (six concierge scenarios) 0.67 #11 of 11 Official leaderboard
VAmoS Pro Bench 25.6% #3 of 3 Official leaderboard
VAmoS Bench 62.3% #1 of 1 Official leaderboard

Vision, audio and robotics

BenchmarkScoreRankSource
Cekura Real Human Calls 25.7% #2 of 7 Official leaderboard
VideoFDB 2.84 #2 of 5 Official leaderboard
Audio MC 36.1% #10 of 21 Official leaderboard
Cekura Turn Bench 82.8% #4 of 8 Official leaderboard
TurnBench 65.7% #1 of 2 Official leaderboard
VoxParity 0.09 #8 of 14 Official leaderboard
Cekura Agent Performance 62.2% #4 of 7 Official leaderboard
Jarvis Bench 981 #3 of 5 Official leaderboard
τ-Voice 43.9% #5 of 8 Official leaderboard
Audio Arena 74.6 #6 of 8 Official leaderboard
EVA-Bench 0.29 #9 of 11 Official leaderboard
People Making Conversation 0.00 #3 of 3 Official leaderboard

What does it cost to run?

On the 2 leaderboards that publish a price, its best run usually costs about the same as the median model's. On 1 of them, no model comes close to its score for less.

Benchmark Score Price From cheapest to priciest Against the median Close to its score for less Close for less
Cekura Agent Performance 62.2% $0.13 1.0× the median 1.0× None
Cekura Real Human Calls 25.7% $0.34 1.2× the median 1.2× Grok Voice 3.0 42.9% for $0.29

Who measured it?

Independent runs cover all 17 of its benchmarks.

  • An independent run only 17
Official leaderboards scored it on 17.

Sources: Official leaderboards, Epoch AI (CC BY 4.0) and labs’ model cards