GOOGLESPEECH-TO-SPEECH1,882 SAMPLES / 30 DAYSLAST RUN AUG 27, 2026, 00:00 UTC

Gemini 3.1 Flash Live (Preview) speech-to-speech benchmarks

Name: Gemini 3.1 Flash Live (Preview) — Google speech-to-speech benchmark results
Creator: Coval

Gemini Live is Google's native-audio path for real-time spoken interaction.

Voice-to-Voice Latency#2 / 2: 1380ms
Instruction Adherence#1 / 2: 76%

Overview

Google's Live API accepts and returns audio in one bidirectional session rather than exposing separate STT and TTS endpoints.

Voice-to-voice latency comes from the call recording instead of provider-internal event timestamps.

Gemini 3.1 Flash Live (Preview) is tested on multi-turn simulated phone calls, measuring the pause before each reply and how closely it follows its instructions.

Technical specifications

Made by: Google
Hosted by: Google
Source: Official API
Licensing: Proprietary
Deployment: Cloud
Region: US
Features: Multilingual

How Gemini 3.1 Flash Live (Preview) ranks

Full S2S dashboard

Voice-to-Voice Latency — every measured S2S modelMilliseconds · lower is better · 30-day average

#1GPT Realtime 21306 ms
#2Gemini 3.1 Flash Live (Preview)1380 ms

median of all models · 1343 ms

S2S models over the last 30 days, ranked on Voice-to-Voice Latency.
#	Model	Host	V2V	Instruction	Samples
1	GPT Realtime 2	OpenAI	1306 ms	71%	838
2	Gemini 3.1 Flash Live (Preview)	Google	1380 ms	76%	834

Under-sampled models are excluded; tied models share a place.

Highest relative placement: 1st of 2 on Instruction Adherence.

Latency vs accuracy

V2V vs InstructionEach point is one measured model · Gemini 3.1 Flash Live (Preview) highlighted · 30-day averages

Averages and tail latency

Averages hide slow outliers — these are the distributions behind each figure.

Gemini 3.1 Flash Live (Preview) latency distributionMilliseconds · shared axis across metrics · last 30 days

Voice-to-Voice Latencyp50 1320 ms · p99 2570 ms

band p25–p75 · tick p50 · whisker to p99 with p90 and p95 stops

Average and percentile values per metric, with the number of samples behind each row.
Metric	Average	p25	p50	p75	p90	p95	p99	Samples
Voice-to-Voice Latency	1380 ms	1200 ms	1320 ms	1460 ms	1633 ms	1775 ms	2570 ms	834
Instruction Adherence	76%	100%	100%	100%	100%	100%	100%	753

Last 30 days

Daily medians from the same measurement runs · gaps are days without qualifying runs.

Voice-to-Voice Latency — daily p50Line p50 · band p25–p75 · UTC days

Instruction Adherence — daily averageDaily average · UTC days

Limits of this comparison

Coval evaluates multi-turn simulated calls, measuring the pause a caller hears and whether the model follows its assigned instructions.

The benchmark covers Coval's published call scenarios and clean-caller condition, not every Gemini capability, tool or language.

Official sources

Google Gemini Live API (documentation)

Results are re-measured daily using fixed datasets and reported over a rolling 30-day window. Same datasets, prompts and metric definitions for every model, measured by Coval’s open-source runner. Full methodology on the overview.

Evaluate your own voice agent

Use Coval to test your production configuration, prompts and calls—not only the public benchmark endpoints.

Book a Demo