Google ships Gemini 3.8 Live and 3.8 Live Extended Thinking, its GA answer to OpenAI's GPT-Live voice models
2026-09-15
Google released two new audio-to-audio models generally available via the Live API: Gemini 3.8 Live, a low-latency default for real-time voice-agent use with interleaved reasoning and asynchronous function calling, and Gemini 3.8 Live Extended Thinking, a higher-reasoning variant for background reasoning during live audio. Simon Willison built and published a working browser voice UI directly from the documentation the same day, confirming the WebSocket API is usable without additional libraries.
Significance 3: A GA release that closes a real-time voice-agent capability gap against OpenAI's GPT-Live family, corroborated independently by a practitioner building against it the same day, but not a frontier reasoning/coding result.
Implications · machine-drafted, not owner judgment
Gives Google a shipped, GA counterpart to OpenAI's GPT-Live and xAI's Grok Voice in the fastest-growing agentic modality (real-time voice), on the same API surface as text Gemini, which matters for developer mindshare in voice agents. Willison's same-day working prototype is a data point that the API is accessible enough for a solo practitioner to build on immediately — directly relevant to what one experienced builder can produce alone with current tools.
- Independent latency/quality comparisons of Gemini Live vs. GPT-Live and Grok Voice
- Enterprise or developer adoption specifically citing Gemini Live within the next month
Sources
- primary Gemini API release notes retrieved 2026-09-16
- primary Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking retrieved 2026-09-16
- curator Gemini Live audio retrieved 2026-09-16