Gemini 3.8 Live
A real-time conversational AI model released by Google, targeting large-scale and cost-efficient deployment scenarios. It integrates conversational intelligence, voice interaction, and visual understanding capabilities, supports automatic switching across 97 languages, and can execute tool calls in the background while maintaining conversational continuity.
Core Facts
The Extended Thinking version of Gemini 3.8 Live incorporates a more thorough thinking process before voice responses, making it suitable for complex reasoning scenarios, while the standard version places greater emphasis on response speed.
Google launched two real-time conversational models, Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking.
Timeline (last 90 days)
Migrating to new models mainly involves changes to model IDs, rather than a rewrite of the entire API architecture.
Developers should first run the complete workflow with the new model in a test environment, then gradually replace the production environment configuration.
During migration, the Live API's session initialization parameters (such as audio sample rate, voice activity detection, and context window configuration fields) and the TTS request body (such as voice selection, output format, and rate parameters) need to be adjusted.
Currently, publicly available information about Gemini 3.8 Live comes only from a single official teaser tweet, lacking technical details, performance benchmarks, context length, supported platforms, and pricing information.
Google officially announced a teaser for Gemini 3.8 Live in the form of a long tweet thread (🧵).
Google released Gemini 3.8 Live with a Live Avatar feature, introducing near-real-time visual rendering capabilities.
Both models support background tool execution, allowing external tools to be called while a conversation is in progress without interrupting the audio stream.
Scenarios prioritizing real-time performance (customer service, voice assistants, real-time translation) should favor Gemini 3.8 Live, while scenarios prioritizing accuracy (complex consulting, educational tutoring, multi-step tasks) should choose the Extended Thinking version.
Developers can access these two models through the Live API.
Google adopts a per-minute pricing model for these two models rather than per-token pricing.
15 more timeline events
All Facts (16)
The Extended Thinking version of Gemini 3.8 Live incorporates a more thorough thinking process before voice responses, making it suitable for complex reasoning scenarios, while the standard version places greater emphasis on response speed.
75%VerifiedGoogle launched two real-time conversational models, Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking.
90%UnverifiedDevelopers should first run the complete workflow with the new model in a test environment, then gradually replace the production environment configuration.
50%UnverifiedMigrating to new models mainly involves changes to model IDs, rather than a rewrite of the entire API architecture.
50%UnverifiedDuring migration, the Live API's session initialization parameters (such as audio sample rate, voice activity detection, and context window configuration fields) and the TTS request body (such as voice selection, output format, and rate parameters) need to be adjusted.
50%UnverifiedGoogle adopts a per-minute pricing model for these two models rather than per-token pricing.
50%UnverifiedDevelopers can access these two models through the Live API.
50%UnverifiedScenarios prioritizing real-time performance (customer service, voice assistants, real-time translation) should favor Gemini 3.8 Live, while scenarios prioritizing accuracy (complex consulting, educational tutoring, multi-step tasks) should choose the Extended Thinking version.
50%UnverifiedBoth models support background tool execution, allowing external tools to be called while a conversation is in progress without interrupting the audio stream.
50%UnverifiedThe browser-based demo interface supports selecting models, switching voice presets, filling in system prompts, initiating voice conversations, and interrupting the model while it is speaking.
50%UnverifiedGemini Live's WebSocket endpoint uses the BidiGenerateContent (bidirectional content generation) method, establishing a persistent bidirectional audio stream between the client and the model.
50%UnverifiedGoogle officially provides a Gemini Live WebSocket tutorial as a starting reference for developer integration.
50%UnverifiedGemini 3.8 Live targets scale and cost-efficiency scenarios, combining conversational intelligence, voice interaction, and visual understanding.
50%UnverifiedThe model can execute tool calls and API requests in the background while keeping the conversation going, using an acknowledge-and-continue non-blocking confirmation mechanism.
50%UnverifiedGoogle released Gemini 3.8 Live with a Live Avatar feature, introducing near-real-time visual rendering capabilities.
70%UnverifiedSimon Willison conducted a quick hands-on test of Gemini 3.8 Live and built a web interface for trying it out in the browser.
50%