Google Nears Release of Flagship Gemini 4 AI Model

Laura Bratton from AI Agenda Live: new Google DeepMind head Koray Kavukcuoglu (first media appearance in role) said Gemini 4 is in early post-training and he hopes release “much earlier” than year-end; intention to ship an early post-training output “as soon as possible” then iterate fast. First next-gen flagship since Gemini 3 late last year (3.1 in Feb; Flash line since). Pichai had previewed Gemini 3.5 Pro for June at I/O—never shipped; Kavukcuoglu said Google “took a little bit of a step back” after 3/3.1 to focus Flash and learning speed; focus now Gemini 4 (3.5 Pro status unclear). Peers ahead: Anthropic Mythos (spring, upgraded this month); OpenAI GPT-6 (early Sep). Kavukcuoglu “a hundred percent confident” Google stays at the frontier. Guardrails/safety testing underway; Gemini 4 used internally for Antigravity coding tool. Leadership: took DeepMind SVP helm in August after Demis Hassabis became chairman/Alphabet chief scientist; also Google chief AI architect. Model work feeds TPU roadmap (“next two or three generations”) as Google sells TPUs to customers vs cloud-only. Caveats: timing is hope/aspiration not a hard date; AGI comments are philosophical; competitive lag acknowledged then waved off.