Koray Kavukcuoglu confirms that Google's next DeepMind model has entered the post-workout phase and hopes for a release well before the end of the year.
Questioned this week at the AI Agenda Live summit, Koray Kavukcuoglu, Technical Director of Google DeepMind, confirmed that Gemini 4 had entered the post-training phase — the stage where the model's behaviour is refined, tested from a safety point of view and supervised by safeguards. His intention, he said, was to publish "as soon as possible" a first result of this post-training, and he hoped for a much earlier availability by the end of 2026.
No date has been communicated. However, the model is already used internally, notably via Antigravity. The calendar of previous versions gives the scale: Gemini 3 Pro in November 2025, Gemini 3.1 Pro in February 2026, then a series of Flash versions, the last of which, Gemini 3.8 Flash, has just arrived.
The context explains the eagerness. Anthropic and OpenAI both delivered major updates in September, and the reproach of a "late" Google comes back regularly. Kavukcuoglu responded by showing confidence in his team rather than promising a date.
Read the full article on the original source
Go to the source →