Google DeepMindNotable

Gemini 3.8 Live

Released Sep 15, 2026

Google's default Live API model for low-latency voice agents, built on Gemini 3 Pro. It takes live audio, images and video and answers in speech, and it also powers Search Live.

Price per minute: —
Pricing: Paid tier, shared with 3.8 Live Extended Thinking and 3.1 Flash Live Preview: input text $0.75 / audio $3.00 (or $0.005/min) / image-video $1.00 (or $0.002/min) per 1M tokens; output text $4.50 / audio $12.00 (or $0.018/min) per 1M tokens, including thinking tokens. Free tier available. Google publishes separate per-minute input and output rates, not one blended per-minute price, so perMinuteUsd is null.
Inputs: audio, text, image, video
Tool calling: Yes
Languages: 97
Voices: 30
Big Bench Audio: 91.7%
Access: Generally available
Licence: Proprietary

Strengths & weaknesses

Strengths

Switches between 97 languages mid-conversation (Google launch post)
Asynchronous, non-blocking function calling plus Google Search grounding
Accepts live video and images as well as audio
Audio at $0.005/min in and $0.018/min out

Weaknesses

Big Bench Audio 91.7%, behind its Extended Thinking sibling's 97.7% (AA)
Affective dialogue removed and proactive audio can't be switched off

Notable: frontier-lab clause (also #9 on AA S2S Index, 74.5). GA 2026-09-15 per the Gemini API changelog. Google's launch post claims second place in AA's Speech Agent Arena. Language count conflicts: the launch post says 97, the Live API guide lists 99, and the Live API overview says 70. Voices: Google Cloud's Live API voice page lists 30 prebuilt voices for the Live API generally, not per model. 131,072 input / 65,536 output tokens. Knowledge cutoff January 2025 (model card).

Links

API model strings

googlegemini-3.8-live