Google DeepMind’s New Chief Pushes Gemini 4 as AGI Talk Takes a Back Seat

Google DeepMind chief Koray Kavukcuoglu says Gemini 4 is in an early phase of post-training, where a base model is refined to behave more reliably. He said he hopes to release it “much earlier” than the end of the year, with fast iterations to follow. Google is still working on safeguards and safety testing, while engineers are already using the model internally in the coding tool Antigravity.

Gemini 4 would be Google’s first new flagship model since Gemini 3 late last year, after Gemini 3.1 and smaller Flash models followed. Sundar Pichai announced Gemini 3.5 Pro at I/O for a June release, but it never arrived, and Kavukcuoglu did not say whether it has been officially canceled. The reported delay matters because OpenAI and Anthropic have continued rolling out competing models while Google prepares its next flagship.

Kavukcuoglu described the question of whether AGI has been achieved as “not the right conversation.” Instead, he said the focus should be on building intelligent agents that can be trusted. That marks a notable tonal shift from Demis Hassabis, who had long centered DeepMind’s identity on the pursuit of artificial general intelligence.

The Decoder frames the shift as the culmination of organizational pressure following ChatGPT, the merger of Google Brain and DeepMind, leadership changes, and departures of prominent researchers to rivals and startups. Kavukcuoglu’s comments suggest Google DeepMind is now prioritizing model release cadence, safety testing, internal tooling, and commercial competition. For readers tracking frontier AI, the key question is whether Gemini 4 can restore Google’s position while redefining DeepMind’s role inside Alphabet.