Gemini 3.8 Live Extended Thinking audio-to-audio

Google released Gemini 3.8 Live Extended Thinking, an audio-to-audio model that performs background reasoning and asynchronous tool calls while streaming continuous audio, with a 131,072-token input limit and 65,536-token output limit. The model, updated September 2026, requires clients to keep listening after turnComplete: true and to track the interaction_status field (IN_PROGRESS or IDLE), and it supports only asynchronous non-blocking function calling. Background reasoning is configured via thinking_config with thinking_level set to low, medium, or high; MINIMAL is not supported.

Gemini 3.8 Live Extended Thinking is our high-reasoning audio-to-audio model recommended when higher background reasoning is required for complex, multi-step problem solving during real-time voice interactions. It processes background reasoning and asynchronous tool calls while streaming continuous audio responses. Documentation Visit the Live API https://ai.google.dev/gemini-api/docs/live-api guide for full coverage of features and capabilities. gemini-3.8-live-extended-thinking | Property | Description | |---|---| | Model code | gemini-3.8-live-extended-thinking | | Supported data types | Inputs Text, images, audio, video Output Text and audio | | Token limits <sup \ \ https://ai.google.dev/gemini-api/docs/tokens </sup | Input token limit 131,072 Output token limit 65,536 | | Capabilities | Supported Not supported Not supported Not supported Supported Async only Not supported Not supported Supported Supported Not supported Supported Not supported | | Consumption options | Not supported | | Versions | model version patterns https://ai.google.dev/gemini-api/docs/models/gemini model-versions for more details. | | Latest update | September 2026 | | Model card | Model card https://deepmind.google/models/model-cards/gemini-3-8-audio/ | Upgrading to Gemini 3.8 Live Extended Thinking Gemini 3.8 Live Extended Thinking introduces background reasoning during live audio sessions. When integrating this model, update your client state management to handle asynchronous reasoning signals: - Asynchronous reasoning protocol : When interacting with models that use asynchronous reasoning, turnComplete: true no longer indicates that the model is idle. The server may continue processing background reasoning or tool calls. Your client must continue listening for subsequent server messages such as tool calls or audio frames after turnComplete: true arrives. - Monitoring interaction status : Use the interaction status field on incoming server messages to determine current server state: - IN PROGRESS : The server is actively processing user input, running background reasoning, or awaiting responses for asynchronous tool calls. Additional model output or tool calls may follow. - IDLE : The server has finished all processing, reasoning, and tool calls. The session is idle and waiting for user input. - Asynchronous function calling : Only asynchronous non-blocking execution behavior: NON BLOCKING is supported. Synchronous blocking mode is not supported and returns a hard error. Function scheduling configurations are not supported. - Thinking configuration : Configure background reasoning using thinking config in your setup configuration thinking level : low , medium , or high . Note that MINIMAL is not supported. - Client content updates : send client content is supported throughout the entire session lifecycle with explicit roles user or model . Setting turn complete=true immediately interrupts active generation. - Proactive audio : Permanently enabled. Setting proactive audio: false returns an error. For a feature comparison across all Live API models, see the Model comparison https://ai.google.dev/gemini-api/docs/live-api/capabilities model-comparison table.