Koray Kavukcuoglu, who took over leadership of Google Deepmind in August following the departure of Demis Hassabis, announced plans to release the Gemini 4 model before the end of the year. The model is currently in the post-training phase, with internal testing already underway via the Antigravity coding tool.
The shift marks a change in strategy for the lab, which previously focused on the pursuit of artificial general intelligence. Kavukcuoglu stated that the AGI debate is not the right conversation, emphasizing instead the development of trustworthy intelligent agents. Development operations are now shifting to the Bay Area.
Google faces pressure to regain momentum after the Gemini 3.5 Pro project missed multiple release deadlines. The company continues to work on safety testing and safeguards for its upcoming flagship release. New models are also intended to inform the design of future generations of Google's custom TPU chips.
The organization is experiencing a significant talent drain, with key researchers departing for competitors like OpenAI and Anthropic. According to Fortune, the ratio of new hires to departures at Deepmind dropped from 12 to 1 in the second quarter of 2023 to 2 to 1 by the third quarter of 2026.
