On September 15, 2026, Google DeepMind launched Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, two new live-dialogue AI models for voice tasks. The models are the team’s most sophisticated live-dialogue systems, offering higher intelligence and parallel reasoning for faster replies and extended deliberation in multi-step voice interactions.
Short answer: On September 15, 2026, Google DeepMind launched Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, two new live-dialogue AI models for voice tasks. The models are the team’s most sophisticated live-dialogue systems, offering higher intelligence and parallel reasoning for faster replies and extended deliberation in multi-step voice interactions.
Google DeepMind dropped the news on September 15, 2026 that two fresh flavors of the Gemini family are now live: Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking. The write-up appeared on the company’s official blog, penned by Tom Ouyang-a principal engineer-and Malini Jaganathan, who sits on the technical staff and works with the Gemini Audio Team. According to the post, these models are the most sophisticated live-dialogue systems the team has ever shipped.
The big leap they’re touting is a noticeable jump in overall smarts paired with a sharper ability to reason in parallel. That combo lets the models glide through tangled conversations and juggle several trains of thought at once. Folks who talk to the system by voice-developers and everyday users alike-should find it smoother to team up on knotty tasks because the AI can hold onto context while it’s simultaneously ticking off different steps of a problem.
Since they’re built for live back-and-forth, the models are tuned to spit out replies fast when they hear spoken input. The extended-thinking variant throws in an extra layer of deliberation, letting the model slice a request into sub-tasks, weigh alternative routes, and polish its answer before it even speaks. That design is meant to cut down the endless ping-pong that often happens when you try to steer an AI through a multi-step process using only voice commands.
The blog stresses that the upgrade isn’t just raw horsepower; it’s about making the whole thing feel more intuitive. By tightening up parallel reasoning, the system can start guessing what you might need next and toss out helpful suggestions without you having to ask. That proactive vibe is aimed at turning the AI into a genuine partner during brainstorming, planning, or problem-solving sessions.
For developers, the release swings open a door to new voice-first apps that hinge on real-time dialogue. The models can slip into products that need hands-free operation-think accessibility tools, wearables, or interactive kiosks. The extended-thinking version shines when you want the AI to explore a few options before locking in a plan, whether you’re drafting a piece of writing, troubleshooting a tech glitch, or organizing a schedule.
Google DeepMind also reminds everyone that the models still follow the company’s safety and responsibility playbook. The live-dialogue architecture includes checks to spot and dampen harmful outputs, and the team ran extensive tests to be sure the boosted reasoning doesn’t sneak in any unintended biases. They invite users to send feedback through the usual channels so future iterations can keep getting better.
All in all, the arrival of Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking nudges us toward more fluid, voice-driven AI interactions. By beefing up both intelligence and parallel reasoning, the models aim to smooth out the friction people often feel when they try to tackle complex goals with spoken language alone. If you’re building AI-powered products or you rely on voice interfaces for work, it’s worth giving these new capabilities a look to see how they could boost responsiveness and usefulness.
Google DeepMind announced the launch of Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking on September 15, 2026, via a post on its official blog.
The announcement was written by Tom Ouyang, a principal engineer at Google DeepMind, and Malini Jaganathan, a member of the technical staff who works with the Gemini Audio Team.
The Extended Thinking version adds an extra deliberation layer that splits requests into sub-tasks, weighs alternative approaches, and refines the answer before speaking, reducing back-and-forth in multi-step voice interactions.
The models enable real-time dialogue for hands-free products such as accessibility tools, wearables, and interactive kiosks, and the Extended Thinking version helps explore options before committing to a plan in tasks like writing, troubleshooting, or scheduling.
Source: Google DeepMind
In May 2026, hundreds of malicious packages uploaded to RubyGems were traced to automated accounts that identified themselves as originating from OpenAI. Researchers said the uploads resembled LLM-generated text, the accounts bypassed email verification, flooded the repository, triggered the build system in an attempt to harvest API keys, and while no successful theft was confirmed, OpenAI acknowledged similar prior activity and has not yet commented.
The New Mexico Supreme Court found defense lawyer Stephen Aarons in direct contempt of court for filing a brief that contained AI-fabricated witness testimony, fined him $5,000, barred him from appearing before the court, and ordered a new lawyer for his client.
New Mexico’s Supreme Court fined defense attorney Stephen Aarons $5,000 and held him in contempt after his appellate brief contained fabricated witness statements and false details about the shooter’s appearance that were generated by ChatGPT.
NO COMMENTS YET
Comments are open. Have a thought or a question? Share it below.