Gemini 3.8 Live Extended Thinking is a voice-oriented variant in Google's Gemini model family, introduced alongside the base Gemini 3.8 Live on September 15, 2026. According to Google's own developer documentation hosted on ai.google.dev, the model is documented under the Gemini API as a distinct entry, signaling that it is treated as a separately addressable endpoint rather than a minor patch to the live audio model. The Jetstream announcement frames the Extended Thinking version as a counterpart tuned for more demanding reasoning, building on the conversational strengths of the live line while pushing toward multi-step processing that voice-only assistants historically struggled to support.
Practically, this variant is aimed at developers who want a single model that can both hold natural spoken dialogue and carry out the chain-of-thought work needed for harder queries, such as stepwise troubleshooting during a voice call or reasoning aloud over a spoken prompt. Its positioning next to Gemini 3.8 Live suggests a design intent of keeping latency and conversational fluency intact while allocating extra budget for internal deliberation, making it a better fit than the base live model for assistants that must reason rather than merely respond. For teams evaluating it, the key fit is voice-first applications where richer thinking is required without sacrificing the real-time feel of the Live family.
