Google DeepMind is in the final refinement stage of its next flagship model, Gemini 4, with plans to launch it well before the end of 2026. The update comes after months of silence on a promised Gemini 3.5 Pro release and a leadership change that saw former AI chief Demis Hassabis step down in August.
A delayed flagship and a shift in strategy
Koray Kavukcuoglu, now leading Google's DeepMind division, made the announcement during his first media appearance in the role. "Our intention is to, like - as soon as possible - to release an early post-training output because we see the results and we are excited," Kavukcuoglu told The Information. He said Google would "continue the fast-paced iterations" going forward.
Google has not shipped a new flagship AI model since the Gemini 3 series launched in November 2025. During that gap, OpenAI released GPT-6 and Anthropic introduced its Mythos series. Both models outperform Gemini 3 on key benchmarks, giving rival AI providers a measurable lead. At the I/O event in May, CEO Sundar Pichai said a major Gemini 3.5 Pro update would arrive in June to help close that gap. The model never rolled out.
Kavukcuoglu explained the decision. The company "took a little bit of a step back" to focus on releasing faster, less powerful Flash models rather than pushing out Gemini 3.5. The move prioritized speed and efficiency over raw capability in the short term.
Leadership transition and what comes next
The delay and subsequent strategy shift unfolded against a backdrop of executive change. Demis Hassabis stepped down from his role in August, saying at the time that "great progress" was being made on Gemini 4. Kavukcuoglu's public comments now signal that the work is nearing completion, with a launch target that he described as "much earlier" than the end of the year.
For teams building on Google's AI stack, the timeline matters. The extended gap between Gemini 3 and Gemini 4 has meant months of operating with a model that trails competitors on performance. A faster release cadence, if sustained, could reduce that disadvantage. Professionals tracking model capabilities for product development or research planning can look to Generative AI Courses that cover evaluation frameworks for comparing frontier models like those from Google, OpenAI, and Anthropic.
Why this matters for product and research teams
The Gemini 4 release will directly affect anyone building on Google's AI infrastructure. A model that closes the performance gap with GPT-6 and Mythos changes the calculus for which provider to bet on. The admission that Google deprioritized a mid-cycle update in favor of faster, lighter models also signals how the company is thinking about the trade-off between capability and deployment speed. Development teams should assess whether their workflows depend on raw model power or can benefit from the Flash-class models Google has been shipping instead. For those evaluating Google's ecosystem, Google AI Courses offer practical grounding in the tools and APIs likely to support Gemini 4 at launch.
Your membership also unlocks:
