Gemini 3.8 Live Extended Thinking Reaches General Availability
Google released Gemini 3.8 Live Extended Thinking on September 15, 2026. This audio-to-audio model supports background reasoning during live interactions via…
Official Release Details
The source event date is September 15, 2026. On this date, Google confirmed that Gemini 3.8 Live Extended Thinking reached general availability (GA). This release occurred alongside Gemini 3.8 Live. Both models are described as new audio-to-audio models intended for real-time voice applications utilizing the Live API.
- Model identifier: gemini-3.8-live-extended-thinking
- Provider: Google
- Status: Generally Available (GA)
- Input modalities: text, image
- Output modality: text
Documented Capabilities and Limits
The official changelog describes Gemini 3.8 Live Extended Thinking as a high-reasoning audio-to-audio model. Its primary documented feature is supporting background reasoning during live audio interactions. The documentation recommends this specific model when higher background reasoning is required. The model has a context window of 131,072 tokens and a maximum output of 65,536 tokens. Open weights are not confirmed for this release. No pricing information is provided in the source facts.
- Supports background reasoning during live audio interactions
- Recommended for scenarios requiring higher background reasoning
- Context window: 131,072 tokens
- Max output tokens: 65,536
- Open weights status: Not confirmed
Evaluation Procedure for Engineers
To evaluate this model, engineers should design tests focused on its stated capability: background reasoning during live audio. A practical procedure involves creating a controlled test environment using the Live API. First, define a set of complex queries that require multi-step logical processing while maintaining a continuous audio stream. Second, measure the latency and accuracy of the responses compared to baseline expectations or previous non-extended thinking variants if available for comparison. Third, validate whether the background reasoning process interferes with the real-time nature of the audio interaction by monitoring interruption rates. Finally, compare the results against the specific requirement for 'higher background reasoning' to determine if the model meets the project's technical needs. These steps allow for an objective assessment without assuming unverified benefits.
Primary sources
Frequently asked questions
What is the difference between Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking?
Both models were released simultaneously on September 15, 2026, as audio-to-audio models for the Live API. The source explicitly states that Gemini 3.8 Live Extended Thinking is a high-reasoning model supporting background reasoning during live audio interactions. It is specifically recommended when higher background reasoning is required. The source does not provide detailed comparative metrics or performance rankings between the two versions.
Can I install Gemini 3.8 Live Extended Thinking locally?
The source facts indicate that open weights are not confirmed for this model. However, the source does not state that local installation is impossible, nor does it confirm that API access is mandatory. The documentation references the Live API guide for getting started, but no definitive statement regarding deployment architecture or local hosting requirements is provided in the verified data.