Google co-founder Sergey Brin told a Stanford audience on December 12 that the publicly available version of Gemini Live is running on an "ancient model" and that users should wait for a major upgrade coming within weeks. During a conversation at the Stanford School of Engineering Centennial celebration, Brin made an unusually candid admission about the gap between what Google employees use internally and what's available to the public.
The admission
Brin's comments came during a discussion about how he uses AI products personally. "I do talk to Gemini Live in the car often," Brin said, before quickly qualifying: "but you shouldn't do it now because we have like way better version coming."
Then the kicker: "The publicly available version right now is not our good version. So like, you shouldn't do it today, but give me a few weeks to actually ship what I have access to, because we have like an ancient model behind it in the publicly research version right now."
He called it "a little embarrassing."
That word choice matters. This isn't a PR rep hedging about competitive positioning. This is Google's co-founder, who came out of retirement specifically to work on Gemini, telling a room full of Stanford engineers and students that the product his company shipped to consumers isn't the product he himself uses.
Awkward timing (or not)
Here's where it gets interesting: on the same day Brin made these comments, Google's official blog announced that Gemini 2.5 Flash Native Audio had started rolling out to Gemini Live and Search Live. The upgrade promises improved function calling, better instruction following, and smoother multi-turn conversations.
Coincidence? Maybe. Or maybe Brin knew the upgrade was already shipping when he spoke. Either way, his "give me a few weeks" timeline aligns with a rollout that's already underway.
Google's blog post claims the updated model achieves a 90% adherence rate to developer instructions (up from 84%) and scores 71.5% on ComplexFuncBench Audio, an evaluation for multi-step function calling. Those numbers are hard to verify independently, but they suggest Google is aware the previous version wasn't cutting it.
What it says about Google's AI strategy
Brin's candor is unusual for a company executive discussing a shipping product. But it also reveals something about Google's internal culture around AI development. Brin returned to daily operations at Google in late 2022 after OpenAI's ChatGPT release, and by his own account spends time in the code, debugging and testing Gemini. At Google I/O in May 2025, he declared that Google "fully intends that Gemini will be the very first AGI."
That's a bold claim. But if your co-founder is publicly telling people not to use your voice assistant because a better version exists internally, the gap between aspiration and execution is showing.
The 71-minute Stanford talk covered broader topics including Brin's Stanford years, Google's history, and advice for entrepreneurs. But this particular moment, captured in the transcript, offers a rare glimpse at how even Google's leadership views the company's public AI products versus what's cooking in the labs.
For users who've been underwhelmed by Gemini Live compared to competing voice assistants, Brin's comments are both validating and frustrating. Yes, the version you're using isn't great. No, you didn't get the good one. But apparently, you will soon.
The Gemini 2.5 Flash Native Audio update is rolling out now in the US, with Google saying the improved voice capabilities will bring more natural responses and better conversational flow to Gemini Live.
