What happened
Google released Gemini 3.8 Live and 3.8 Live Extended Thinking, two speech-to-speech models similar in shape to OpenAI's GPT-Live family. A developer built a browser UI letting you pick a model and voice preset, add an optional system prompt, and interrupt the model mid-speech.
Why it matters
The demo shows these new models can be tried in a browser, including interrupting the model while it talks, so the shape of Google's speech-to-speech lineup is now comparable to OpenAI's GPT-Live family.
What to watch
The demo relies on audio capture and playback in the browser, so the experience hinges on whether that works reliably. Watch how the two models differ once the Extended Thinking option is compared in use.
WHO IT HITSDevelopers and product teams evaluating speech-to-speech models can now try Google's new Gemini Live releases through a browser demo, without building their own interface.
Summaries like this, every morning.
Google's release of Gemini 3.8 Live and 3.8 Live Extended Thinking puts two speech-to-speech models into the same general shape as OpenAI's GPT-Live family. That parallel is the body's own framing, not an outside benchmark.
The body also shows how quickly these models can be put in front of people. A developer pointed GPT-6 Astra Extra High at the documentation and had it build a browser UI for trying the new models, with model and voice preset selection, an optional system prompt, and the ability to interrupt the model while it is talking. The implementation connects directly without libraries and uses browser audio APIs for capture and playback.
The stakes for this release hinge on how the two Gemini Live variants compare in practice, especially the Extended Thinking option, and on whether the browser-based route holds up for people who want to test voice interactions themselves. That is likely to matter most to developers and product teams weighing speech-to-speech options, though the body does not report usage numbers or independent evaluations.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
A Dell Technologies survey of 2,950 business and IT decision-makers in 35 countries found 58% of Japanese comp…
CData's guide walks through creating a SELECT-only Snowflake role called claude_role, setting up an OAuth secu…
Oracle announced on Oct 28, 2026 an expanded partnership with OpenAI and a new product lineup spanning four ar…
Mercari's fiscal 2026 June-period earnings material shows all employees were using AI tools by the end of Janu…
Meta's personal AI agent Muse started in Japan on October 23, and unlike text generators it can operate websit…
Flamerix AI analyzed OpenTelemetry exports and found Claude Code's agent summary feature consumes roughly 15-2…
