From the source
Google DeepMind released two generally available Live API models for real-time multimodal dialogue and agentic tool use.
- Gemini 3.8 Live is engineered for sub-second conversational latency and gains native audio understanding, affective dialogue, proactive audio, and synchronous or asynchronous function calling.
- Gemini 3.8 Live Extended Thinking targets complex reasoning, multi-step planning and long-horizon state tracking, with configurable thinking levels of MINIMAL, LOW, MEDIUM and HIGH and asynchronous function calling.
- Documentation for both is available in the Thinking in the Live API guide and the Live API overview.



