Gemini 3.8 Live brings real-time audio agents to the Live API

Google has made Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking generally available through the Gemini Live API. Both are audio-to-audio models for real-time voice applications, but they target different latency and reasoning needs.

Two GA audio-to-audio models

The default model ID is gemini-3.8-live. Google describes it as the option for most low-latency voice-agent experiences and real-time dialogue without reasoning delays. It supports interleaved reasoning, default asynchronous function calling, and full session client-content updates.

The higher-reasoning model is gemini-3.8-live-extended-thinking, which performs background reasoning during live audio interactions. Google recommends it when an application needs more reasoning depth while the conversation continues.

Both models are used with the Live API and are intended for applications where audio input and audio output need to stay connected in real time. Developers should measure turn-taking, interruption handling, function-call latency, and audio quality on their own workloads rather than assuming that a higher reasoning setting is always the better voice experience.

What developers should test

A voice agent still needs explicit boundaries around tool use. Async function calling can keep a conversation moving, but an application should authenticate and validate every consequential tool call, surface state changes to the user, and log the model’s request and the tool’s result.

Teams comparing the two models should test the same prompts with identical tool definitions and session settings. Low-latency dialogue may favor gemini-3.8-live; background reasoning may be worth the extra complexity for multi-step workflows where the agent must plan while the user keeps speaking.

Availability

The Gemini API changelog marks both models GA as of September 15, 2026. The entry does not list additional limitations, quotas, or pricing in the release note, so teams should check the current Live API documentation and account tier before estimating production capacity.

Sources

• Google AI for Developers — Gemini API release notes (September 15, 2026): https://ai.google.dev/gemini-api/docs/changelog
• Google AI for Developers — Live API: https://ai.google.dev/gemini-api/docs/live
• Google for Developers on X — Gemini 3.8 Live announcement (September 15, 2026): https://x.com/googledevs/status/2099908204549624254
• Google for Developers on X — Gemini 3.8 Live Extended Thinking details: https://x.com/googledevs/status/2099908702929342719