Google DeepMind: Kavukcuoglu Aims for the Top, Bets on Flash

Koray Kavukcuoglu, the new head of Google DeepMind, asserts that leading at the forefront is the top priority. He admits that Google's models are still somewhat behind, without providing any new details on Gemini 4 or Gemini 3.5 Pro, which apparently remain in development and are several months delayed. He highlights the rapid progress of the Flash series, described as a shift towards coding agents, and emphasizes the central role of software engineering.

Flash 3.5→3.7: Transition to a Coding Agent

Koray Kavukcuoglu emphasizes the rapid advancements made with the Flash series. He describes the transition from Flash 3.5 to 3.6 and then to 3.7 as an evolution from a language model to a coding agent. According to him, the goal is to transform the model into an agent. He also believes that software engineering is the most critical domain and environment for the success of these systems.

Gemini 4 with No Public Updates, 3.5 Pro Still Delayed

Regarding Gemini 4, Koray Kavukcuoglu reiterates that it is the most ambitious race to date and notes that, knock on wood, everything is going well. However, no concrete new information has been shared. No updates have been provided on Gemini 3.5 Pro, which still seems to be in development and is now several months behind schedule.

Asserted Leadership Despite Acknowledged Gap with the Forefront

Koray Kavukcuoglu, recently appointed to lead Google DeepMind, claims that the only thing that matters is being at the forefront and expresses 100% confidence that the company will achieve this. However, he acknowledges that Google's current models are somewhat below that level. He assures that the team, resources, and technology needed are in place to close this gap. These statements come as some have speculated that Google may not be seeking to lead the AI model race, prioritizing cost-effectiveness instead.