Google releases Gemini 3.8 Live speech models to match OpenAI's voice capability
Source published ·Modelwire updated
Original coverage: Simon Willison ↗·How Modelwire adds context

The development
Google shipped Gemini 3.8 Live and 3.8 Live Extended Thinking, speech-to-speech models that directly mirror OpenAI's GPT-Live architecture. This marks Google's competitive response to real-time voice interaction, a capability gap that has favored OpenAI since GPT-Live's launch. The release signals intensifying parity in conversational AI, with both labs now offering low-latency voice reasoning. Simon Willison built a web UI demonstrating the models' accessibility to developers, lowering friction for adoption and experimentation across the ecosystem.
Modelwire’s AI-generated summary of coverage from Simon Willison.
Modelwire analysis
Analyst takeOur AI-generated reading of the wider context and the next developments to watch.
The more consequential detail here is not the model release itself but the naming convention: Google calling these '3.8 Live' signals a versioning strategy that ties voice capability directly to its frontier model line, rather than treating voice as a separate product tier the way OpenAI initially did with GPT-4o's phased rollout.
This is largely disconnected from recent activity in our archive, as we have no prior coverage to anchor against. In the broader competitive context, though, this belongs to a pattern that has been building since OpenAI first demonstrated low-latency speech-to-speech in mid-2024: each lab has treated real-time voice as a prestige capability, and Google's gap on that front has been a recurring talking point in developer communities. Willison's decision to build a public demo UI matters here because developer friction is often where parity claims get tested in practice, not in benchmarks.
Watch whether third-party developers report latency and interruption-handling on par with GPT-Live in real applications over the next 60 days. If the gap holds in practice despite the architectural mirroring, that suggests Google's infrastructure, not its models, is the remaining constraint.
This interpretation is generated from the summary above and available source metadata. Our methodology · Report an error
MentionsGoogle · Gemini 3.8 Live · Gemini 3.8 Live Extended Thinking · OpenAI · GPT-Live · Simon Willison
How this coverage is produced
Modelwire uses AI to generate summaries and context from source headlines, snippets, and selected archive coverage. Automated checks do not verify every claim, and items are not routinely reviewed by a person before publication. Zacaria Solis operates the site. Read the linked source for the full evidence and report errors through our corrections process.
Modelwire summarizes, we don’t republish. Simon Willison originally reported this story as “Gemini Live audio”. The full content lives on simonwillison.net. If you’re a publisher and want a different summarization policy for your work, see our takedown page.