Gemini 3.8 Live Extended Thinking
Gemini 3.8 Live Extended Thinking is Google's September 15, 2026 high-reasoning Live API model for complex multi-step work during a live voice session.
Provider
Model family
Google Gemini Live
Full-duplex voice model
Cost tier
Live Extended Thinking
Status
Current
Release Sep 15, 2026
Why teams choose it
Function calling is async-only (NON_BLOCKING)
Synchronous BLOCKING returns a hard error. Function scheduling configs are not supported.
turnComplete: true does not mean idle. Watch interaction_status IN_PROGRESS vs IDLE.
thinking_config supports low, medium, high — not MINIMAL.
Google lists the same $0.005/$0.018 per-minute audio rates as 3.8 Live
Extended Thinking is the Gemini Live / Workspace Live SKU, not the default Search Live workhorse.
Tradeoffs to know
- Same Live API gaps as 3.8 Live: no caching, code execution, structured outputs, or image generation on the model card.
- Vendor Speech-to-Speech and τ-Voice scores are Google-cited; reproduce on your own live harness before replacing 3.8 Live.
When not to use this
- Do not use this as a drop-in for Gemini 3.8 Flash coding routes or for ASR-only jobs (use Gemini 3.5 Transcribe).
Benchmarks
- Tau Voice
- 68.6
- Tau Voice Banking
- 35.1
Technical specs
- Inputs
- text, image, audio, video
- Outputs
- audio, text
- License
- Proprietary API
- API ID
gemini-3.8-live-extended-thinking- Context tokens
- 131,072
- Big Bench Audio
- 97.7
- Max output tokens
- 65,536
- Audio input (USD / min)
- 0.005
- Audio output (USD / min)
- 0.018
- Aa Speech To Speech Quality Index
- 82.6
- Capabilities
- full-duplex voice
- background reasoning
- asynchronous function calling
- progress narration
- visual grounding
- search grounding
- audio generation
- configurable thinking
Google Gemini Live family lineup
Current models
Related models
Gemini 3.8 Live
Current · 3.8 Live · live
Gemma 2 27B
Gemma 2 27B is Google’s open-weights Gemma family checkpoint balancing quality and deployability for research and pro…
Gemini 3.1 Pro Preview
Gemini 3.1 Pro Preview is Google's preview Gemini 3-series model for complex multimodal reasoning, software engineeri…
Gemini 3.8 Live Extended Thinking FAQ
What is Gemini 3.8 Live Extended Thinking?
Gemini 3.8 Live Extended Thinking is Google's September 15, 2026 high-reasoning Live API model for complex multi-step work during a live voice session. The stable model code is gemini-3.8-live-extended-thinking. Same 131,072 / 65,536 token limits and text/image/audio/video in, text and audio out as Gemini 3.8 Live. It runs background reasoning and asynchronous tool calls while streaming audio. Google cites #1 on Artificial Analysis Speech-to-Speech Quality Index (82.6), 68.6% on τ-Voice, 35.1% on Sierra τ-Voice-banking, and 97.7% on Big Bench Audio. Same listed audio session rates as 3.8 Live.
When does Gemini 3.8 Live Extended Thinking fit best?
Voice agents that must reason and call tools without stopping speech
What should teams watch out for with Gemini 3.8 Live Extended Thinking?
Do not use this as a drop-in for Gemini 3.8 Flash coding routes or for ASR-only jobs (use Gemini 3.5 Transcribe).
Explore next
Models, tools, and comparisons that connect to this reference.