DATASHEET
Google DeepMind/RELEASED Sep 15, 2026
Gemini 3.8 Live Extended Thinking
#1 SPEECH-TO-SPEECHReleased Sept 15, 2026; Google's native speech-to-speech model that reasons and talks at once, with background tools, 97-language switching, and #1 82.6 on the Speech-to-Speech Quality Index.
Context
131,072 tokens
Architecture
Undisclosed (Gemini 3 native audio stack)
Official API (1M)
$0.75 in / $4.5 out
License
Google AI Studio / Vertex AI
Shipped Variants
Gemini 3.8 Live Extended ThinkingREASONING LIVE
High-complexity live reasoning with configurable thinking (low/medium/high), simultaneous speech and background multi-step tool use for production voice agents.
$0.75 in / $4.50 out per 1M text tokens ($0.005/min in / $0.018/min out audio; thinking billed as output)
OFFICIAL DOCSGemini 3.8 LiveLOW-LATENCY LIVE
Cost-efficient native speech-to-speech for fluid dialogue and near real-time visual grounding with background async tools; #2 Speech Agent Arena.
$0.75 in / $4.50 out per 1M text tokens ($0.005/min in / $0.018/min out audio)
OFFICIAL DOCSQuickstart Snippet
curl -s https://modelregistry.tirup.in/api/cli?model=gemini-3-8-live-extended-thinking