Google's DeepMind division head, Koray Kavukcuoglu, stated Wednesday that the company is nearing the release of its latest flagship model, Gemini 4, a highly anticipated development after the tech giant fell behind rivals Anthropic and OpenAI in the artificial intelligence race. Kavukcuoglu shared at an event that Gemini 4 is in the early stages of the "post-training" phase, where the base AI model is refined to ensure reliable performance, and expressed hope the release would come "well before the end of the year."

"Our intention is to release the early version from the post-training phase as soon as possible, because we are seeing results and feeling excited about them," he said, noting that Google would then "continue with rapid iterations." Gemini 4 will be Google's first next-generation flagship model since the Gemini 3 series launched late last year, though the company did release a Gemini 3.1 update in February along with a range of smaller, cost-effective Flash models that offer weaker complex reasoning capabilities.

At the annual I/O developer conference in May, Google CEO Sundar Pichai had announced a major update, Gemini 3.5 Pro, slated for June, but that version ultimately failed to appear as scheduled. Kavukcuoglu explained that after releasing the Gemini 3 and 3.1 models, the company chose to "slow down slightly" and instead focus on launching the Flash series rather than proceeding with Gemini 3.5 Pro. "For us, the priority at that time was to maximize learning speed," he said, without confirming whether Google has officially abandoned Gemini 3.5 Pro, but making it clear the current focus is squarely on delivering Gemini 4.

Rivals have surged ahead in the interim — Anthropic launched its powerful new Mythos model series this spring and upgraded it this month, while OpenAI began rolling out its equally capable GPT-6 model in early September. Both companies' models outperform Google's current most advanced offering. When asked whether he is concerned that Google has fallen behind its peers, Kavukcuoglu expressed strong confidence in his team. "We have an extremely strong team, and I have full confidence in them," he said. "In my view, we will always be at the forefront of technology."

Kavukcuoglu noted that Google is developing safeguards and conducting safety testing for Gemini 4, and engineers are already using the model internally to run its Antigravity AI coding tool. He assumed leadership of Google's AI operations in August, transitioning from chief technology officer to senior vice president of Google DeepMind, following Demis Hassabis's move from CEO to chairman of the unit and chief scientist at parent company Alphabet. Kavukcuoglu also serves as Google's chief AI architect.

Google has a track record of initially lagging in the AI model race before catching up. Its Gemini 3 model represented a major technological leap that triggered a "red alert" at OpenAI, prompting the company to rush to match the search giant. Kavukcuoglu said that, unlike some peers in the AI industry, he holds no strong opinion on whether the field has achieved AGI (artificial general intelligence). "The discussion about whether we have achieved AGI is not really the important topic," he said. "The more meaningful question is: can we build trustworthy agents?"

Kavukcuoglu added that progress on frontier models will also help drive Google's rapidly growing hardware business, which produces AI chips known as TPUs (tensor processing units). In an effort to strengthen competition with Nvidia, Google began selling TPUs directly to customers this year, rather than only offering access to the hardware through its cloud services. He pointed out that Google DeepMind's research on new models provides critical guidance for hardware engineers as they plan the design direction for the "next two to three generations" of chips.