Google’s latest AI push is not just about Gemini 3 Pro. For Demis Hassabis, CEO of Google DeepMind, the new model is an important step toward broader AI systems, but not the endpoint. The bigger story is what comes next: world models, harder reliability problems, and a market that may be moving faster than the technology can justify.
Gemini 3 Pro targets broader AI leadership
After a hype phase led by Google employees and CEO Sundar Pichai, Google has released Gemini 3 Pro. Hassabis described the aim as building the best all-around model while improving areas where earlier versions were weaker, including coding, reasoning, and math.
In an interview with Hassabis, the model was described as a "key component of what we see as an eventual AGI system." That framing matters. Google is positioning Gemini 3 Pro not only as a product release, but as one part of a longer technical path.
Early reports point to top-tier performance. Josh Woodward, a vice president at Google, said on the "Hard Fork" podcast that Gemini 3 is better at keeping its train of thought across multiple steps. He also highlighted its ability to create custom interfaces for users, including interactive tutorials and calculators.
The scale of distribution is already large. According to Google, the Gemini app has over 650 million monthly users. Google also says two billion people interact with Gemini monthly through AI Overviews in Search. To accelerate adoption further, all U.S. college students will receive one year of free access to a paid version of Gemini.
World models are becoming the next frontier
Even as Gemini 3 Pro rolls out, Hassabis is spending much of his research attention elsewhere. His focus is on world models, which he sees as essential for future AI systems and critical to achieving AGI.
He pointed to SIMA 2 and the video-generation model Genie 3 as examples of this direction. These models are already being used internally for training robots and other agents. The practical idea is that AI systems need richer ways to model environments, actions, and consequences, especially when they are expected to operate beyond text-only tasks.
Hassabis expects world models to have a "ChatGPT moment," but he also made clear that the path is not simple. Cost is one barrier. "We’d love to put Genie in the hands of more people, but it’s expensive," he said. He explained the economics by saying that "basically, a consumer of it is another instance of the creation of it."
There are also technical limits. One challenge he named is "making it consistent longer than a minute." That matters because a system that can only remain coherent briefly is not yet ready for broader use in complex environments, training workflows, or interactive applications.
The AI bubble warning is aimed at private markets
Hassabis offered a careful view of the AI market. He said there is "obviously a bubble in the private market." His example was seed funding rounds where startups with "basically nothing" receive valuations of tens of billions of dollars.
His assessment was direct: "That seems a little unsustainable," he said, adding, "It’s not quite logical to me." The concern is not that AI has no value. It is that some private market valuations appear disconnected from what companies have actually built.
He separated that from Google’s position. Hassabis said his thesis when DeepMind joined Google twelve years ago was that AI research should become the "engine room" for Google’s products. In his view, that strategy is now producing returns because Gemini is already powering major services.
Those services include Search, YouTube, and Cloud. That gives Google a different profile from startups that depend on future promise alone. "Whatever happens, I feel like we’re in a really strong position to come out on top," Hassabis said. Whether or not the bubble bursts, he believes Google is positioned well.
AGI still requires more than bigger models
Gemini 3 Pro does not change Hassabis’s timeline for AGI. He still estimates that true AGI is five to ten years away. He also said "one or two more breakthroughs" are needed, especially in reliability, reasoning, and memory.
That view pushes back against a simple scaling story. Hassabis acknowledged "diminishing returns" from only making models larger, but he did not say progress is ending. His argument is that progress can continue even if it no longer grows exponentially.
People when they hear diminishing returns they think of is it zero or exponential, right? But there’s also in between. So there can be diminishing. It's not like going to like exponentially double with every era, but it’s not um uh it’s but it’s still well worth doing, right? And and and extremely good return on that investment. So, I think we're in that era.
This middle ground is important for understanding Google’s strategy. The company can keep investing in larger and better models while also seeking the breakthroughs that scaling alone may not deliver.
More capable systems bring more risk
Hassabis also pointed to safety concerns as AI systems gain new abilities. He specifically mentioned the use of external tools, known as "function calling," as an area that creates new risks.
Cybersecurity was the clearest example he raised. As models become more capable and more connected to tools, he said one must be "doubly cautious" to prevent misuse. That warning sits alongside the larger message of the Gemini 3 Pro launch: more capability is valuable, but it also raises the stakes for reliability, control, and deployment.
The result is a mixed but clear picture. Gemini 3 Pro strengthens Google’s current AI offering. World models point to the next expensive research phase. AGI remains years away by Hassabis’s estimate. And while the private AI market may be overheated, Google’s bet is that embedding AI deeply into existing products gives it a stronger foundation than hype alone.