Google DeepMind Head Confirms Gemini 4 Enters Post-Training Phase

Google DeepMind's new head, Koray Kavukcuoglu, announced that the next-generation Gemini 4 model has entered the post-training phase. During the AI Agenda Live summit, Kavukcuoglu stated that the company aims to release the model well ahead of the end of 2026. This development follows a ten-month period without a new flagship model release from Google, during which competitors like OpenAI and Anthropic have introduced advanced models. Kavukcuoglu emphasized a strategy of rapid iteration and focus on building reliable intelligent agents rather than prioritizing AGI milestones. Despite the delay in flagship releases, Google reports that its Antigravity platform maintains over 2.4 million weekly active users, with its model APIs processing approximately 22 billion tokens per minute.