Google DeepMind Signals Gemini 4 Launch Well Before Year-End, Targets Rivals' Newer Models

Google DeepMind chief Koray Kavukcuoglu said the company is nearing the release of its next flagship AI model, Gemini 4, and hopes to launch it "well before the end of the year." Early post-training versions will ship first, followed by rapid updates. Kavukcuoglu said Gemini 4, the first next-generation flagship since the Gemini 3 series launched in late 2025, is in early post-training, focused on improving reliability. After Gemini 3 and 3.1, the team slowed down and prioritized cheaper Flash models to speed learning; he did not say whether the delayed Gemini 3.5 Pro, promised for June, has been abandoned. Competitors Anthropic and OpenAI have released stronger models, including OpenAI's GPT-6 in early September. He said Gemini 4 is undergoing safety testing, engineers are already using it internally with the Antigravity coding tool, and frontier-model work will inform the design of Google's next two to three TPU chip generations.