Google DeepMind CEO Demis Hassabis Reflects on Gemini 3's Launch and OpenAI's 'Code Red' Reaction
In a revealing interview with Alex Heath, Google DeepMind CEO Demis Hassabis has shared insights into the launch of Gemini 3, an event that sent shockwaves through the AI industry and prompted OpenAI CEO Sam Altman to declare a company-wide "Code Red" in December. Hassabis admitted that even he was uncertain about how the new model would perform against competitors, underscoring the unpredictable nature of AI development.
The Unpredictable Nature of AI Model Development
"You never quite know when you're baking a new model, right? You obviously have your theories and your tests," Hassabis told Heath. Despite this uncertainty, he emphasized Google's strong trajectory, noting that Gemini 2.5 had already topped leaderboards in May, setting the stage for Gemini 3's impactful release.
OpenAI's Swift User Loss and Strategic Pivot
The fallout from Gemini 3's November release was immediate and significant. According to SimilarWeb data, OpenAI lost approximately 6% of its users within a week, with ChatGPT's daily traffic dropping from 203 million to 191 million average visits. This translated to a loss of roughly 12 million daily visitors, highlighting the competitive pressure.
In response, Altman issued an internal memo instructing staff to pause work on several projects, including advertising, AI health agents, and a personal assistant called Pulse. The focus shifted entirely to improving ChatGPT's speed, reliability, and personalization, a move that Hassabis acknowledged as part of the "ferocious, intense competition" in the AI space.
Google's Secret Weapon: The Pre-Training Team
When asked about Google's competitive edge, Hassabis pointed to the company's pre-training team as a key asset. "When it comes to pre-training, I think we have the best team in the world by far," he stated. He credited Google DeepMind's depth in fundamental research for enabling the continuous delivery of competitive models, which has driven significant growth in user adoption.
The Gemini app's monthly active users surged from 350 million in March to 650 million by October, while the Nano Banana image generator, launched in August, gained such popularity that it nearly overwhelmed Google's tensor processing units.
Looking Ahead: AGI and the Ongoing AI Race
Despite Gemini 3's success, Hassabis remains cautious, noting that Google's generative AI market share has grown from 5% to 20% in a year, but "there's still a lot of hard work in front of us." He subtly criticized competitors raising massive funding rounds without substantial research or products, calling such ventures "very fragile."
Currently, Hassabis and his team are focused on developing Gemini 4, with hopes to accelerate progress further this year. As he aptly summarized, "Everyone's trying to leapfrog each other. So one can't rest on our laurels," indicating that the AI race is far from settled and will continue to evolve with intense rivalry and innovation.
