Google DeepMind chief Koray Kavukcuoglu said Gemini 4 has entered the early stage of post-training, when a foundational AI model is fine-tuned for more consistent and reliable real-world performance. Google intends to release an initial post-training version as soon as possible and continue rapid iterations, with Kavukcuoglu indicating a launch well before the end of this year, although no specific date has been set. The development is closely watched as Google faces stronger competition from Anthropic and OpenAI. The comments were Kavukcuoglu's first substantive public remarks since taking over DeepMind after Demis Hassabis departed his position last August. Alphabet shares fell 3.80% after the remarks, while Meta Platforms rose 1.02%. Google's Gemini 3 debuted in November 2025, but industry assessments later said OpenAI's GPT-6 and Anthropic's Mythos product line had surpassed it. Google also abandoned a previously promised Gemini 3.5 Pro enhancement by June, choosing instead to focus on smaller, more efficient Flash variants. Kavukcuoglu said he considers it certain that Google will remain at the frontier, while Meta reported accelerating adoption of its AI capabilities through September.